JP4935329B2 - Speech coding apparatus, speech decoding apparatus, speech coding method, speech decoding method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce an information amount which is transmitted and received, by maintaining sound quality in voice encoding and decoding of an analysis and synthesis type. <P>SOLUTION: A voice encoding device expresses a residual signal created by applying prediction analysis to a voice signal, by a flag for indicating a noise band for the noise band, and by a frequency conversion coefficient for a non-noise band. The voice encoding device encodes the flag and the frequency conversion coefficient together with a prediction coefficient, and transmits it to a voice decoding device. The voice decoding device decodes a received code, and on the basis of a decoding result, a residual signal is restored by creation of a noise string and reverse conversion of the frequency conversion coefficient, and the voice signal is restored by inputting the residual signal to a filter section for synthesis, which is defined by the prediction coefficient. <P>COPYRIGHT: (C)2008,JPO&amp;INPIT

Description

  The present invention relates to a speech encoding device, speech decoding device, speech encoding method, speech decoding method, and program that are required when performing analysis / synthesis speech compression / decompression.

  In the field of mobile communications, a low bit rate (about 8 kbps) audio compression coding method is required for the purpose of dealing with an increase in users. For example, as an 8 kbps speech coding method, there is a speech coding method shown in ITU-T recommendation G.729. The speech coding method according to the recommendation is basically a method of transmitting a speech signal after decomposing it into a prediction coefficient and a residual signal by predictive analysis. As prediction analysis, for example, linear prediction analysis and MLSA (Mel Log Spectrum Approximation) analysis (for example, refer to Non-Patent Document 1) are known.

Sei Imai, Kazuo Sumita, Chieko Furuichi, “Mel Log Spectrum Approximation (MLSA) Filter for Speech Synthesis”, IEICE Transactions, Vol. J66-A, No. 2, p. 122-129, 1983

  In order to enable the above-described low bit rate communication in the analysis and synthesis type encoding and decoding, it is necessary to devise how to efficiently transmit information on the residual signal from the encoding device side to the decoding device side. Is required. This is because the residual signal has a large amount of information and thus overflows if transmitted as it is.

  As one of such contrivances, a codebook in which typical residual signals are collected is shared between the encoding side device (transmitter) and the decoding side device (receiver), and the former device is directed to the latter device. Thus, it is performed to tell which residual signal is selected from the codebook. However, when a dynamic codebook is used to improve sound quality, there is a problem that a codebook mismatch may occur between the transmitter and the receiver.

  The residual signal generated by the transmitter is used as an excitation signal for sound reproduction in the receiver. In other words, the term “residual signal” and the term “excitation signal” are used differently depending on the situation, and are the same in substance. Therefore, in the following, the usage of both terms is not particular.

  As a device for efficiently transmitting information about the residual signal without using the above-described codebook, it is conceivable that the transmitter performs frequency conversion of the residual signal and then encodes it.

  However, for example, when the voice is a consonant, the residual signal often has a characteristic as noise, and it is wasteful to uniformly transmit the frequency conversion coefficient to the receiver until such time. . That is, the limited transmission capacity cannot be used efficiently.

  The present invention has been made in view of the above circumstances, and in a speech compression decoding technique, a residual signal is divided into a plurality of bands, and a noise band is mainly transmitted with a flag indicating that as a non-noise band. In the case of a speech coding apparatus, a speech decoding apparatus, a speech coding, and the like, the frequency conversion coefficient is set as a transmission target so that the quality of the reproduced speech is sufficiently ensured while efficiently transmitting information on the residual signal. It is an object to provide a method, a speech decoding method, and a program.

In order to achieve the above object, a speech encoding apparatus according to the first aspect of the present invention provides:
A prediction analysis unit that decomposes a speech signal into a prediction coefficient and a residual signal by prediction analysis;
A residual signal generator for each band that divides the residual signal into residual signals for each band;
A noise discriminating unit that discriminates whether or not the band is a noise band for each band of the residual signal;
A flag generating unit for generating a flag indicating that the band is a noise band for the band determined to be a noise band by the noise determining unit and obtaining a gain of a residual signal for each band of the band ;
Non-noise band conversion for generating a frequency conversion coefficient in a non-noise band by superimposing the band-specific residual signals of the band determined not to be a noise band by the noise determination unit in the real-time domain and then performing frequency conversion And
An encoding unit that encodes the prediction coefficient obtained by the prediction analysis unit, the flag and gain obtained by the flag generation unit, and the frequency conversion coefficient generated by the non-noise band conversion unit ;
Is provided.

  The residual signal is divided into a plurality of bands, and for the noise band, only the flag indicating that in principle is to be transmitted, and for the non-noise band, the frequency conversion coefficient is to be transmitted. It is possible to achieve both efficient transmission and ensuring the quality of reproduced audio.

In order to achieve the above object, a speech encoding apparatus according to the second aspect of the present invention provides:
A prediction analysis unit that decomposes a speech signal into a prediction coefficient and a residual signal by prediction analysis;
A full-band conversion unit that frequency-converts the residual signal to generate a frequency conversion coefficient;
A residual signal generator for each band that divides the residual signal into residual signals for each band;
A noise discriminating unit that discriminates whether or not the band is a noise band for each band of the residual signal;
A flag generating unit for generating a flag indicating that the band is a noise band for the band determined to be a noise band by the noise determining unit and obtaining a gain of a residual signal for each band of the band ;
From the frequency conversion coefficients obtained by the all-band conversion unit, a totaling unit that totalizes the frequency conversion coefficients of the band determined not to be a noise band by the noise determination unit,
An encoding unit that encodes the prediction coefficient obtained by the prediction analysis unit, the flag and gain obtained by the flag generation unit, and the frequency conversion coefficient tabulated by the tabulation unit ;
Is provided.

  Since the frequency conversion coefficient is obtained all over the entire band before dividing the residual signal into band-specific residual signals, the frequency conversion coefficient can be obtained with high accuracy regardless of the performance such as the resolution of the band filter. .

  The noise determination unit determines, for example, whether or not the band is a noise band based on the shape of the autocorrelation function of the band-specific residual signal for each band.

  In this way, as will be described in detail later, voiced / unvoiced discrimination can be easily performed by adopting a predetermined standard.

  The prediction analysis unit obtains an MLSA filter coefficient as the prediction coefficient by, for example, MLSA (Mel Log Spectrum Approximation) analysis, and obtains the residual signal using an inverse filter defined by the MLSA filter coefficient.

  The prediction analysis unit obtains a linear prediction coefficient as the prediction coefficient by, for example, linear prediction analysis, and obtains the residual signal using an inverse filter defined by the linear prediction coefficient.

In order to achieve the above object, a speech decoding apparatus according to the third aspect of the present invention provides:
Prediction coefficient generated from speech signal by prediction analysis, flag indicating that specific band of residual signal generated from speech signal by prediction analysis is noise band, gain of residual signal by band in noise band And a receiving unit that receives a code that is a result of encoding a frequency conversion coefficient in a non-noise band ;
A decoding unit that decodes the prediction coefficient, the flag, the gain, and the frequency transform coefficient in the non-noise band from the code;
A noise string generation unit that generates a noise string whose amplitude is adjusted by the gain for each band in a band indicated by the flag to be a noise band;
In the band indicated by the flag as a noise band, all the frequency conversion coefficients in the band are stored as 0, and in the non-noise band, the decoded frequency conversion coefficient is stored to store the entire band. An inverse transform unit that generates a frequency transform coefficient and inversely transforms the generated frequency transform coefficient to obtain a residual signal in a non-noise band ;
A residual signal restoration unit that generates a restored residual signal by superimposing a noise sequence generated by the noise sequence generation unit and a residual signal in a non-noise band obtained by the inverse transformation unit ;
A synthesizing unit that generates a restored speech signal by synthesizing the prediction coefficient decoded by the decoding unit and the restored residual signal generated by the residual signal restoring unit ;
Is provided.

  According to the present invention, it is possible to improve the sound quality of reproduced sound while suppressing the amount of information transmitted in analysis / synthesis type audio signal communication.

  The speech encoding apparatus and speech decoding apparatus according to embodiments of the present invention will be described in detail below. Three types of functional configuration examples of speech encoding apparatuses (respectively, Embodiment 1, Embodiment 2, and Embodiment 3) and two types of functional configuration examples of speech decoding apparatuses (Embodiments 4, 5 respectively). Furthermore, a speech encoding / decoding device (referred to as Embodiment 6) is given as an example of a physical configuration for realizing these functional configuration examples. In addition, in describing these embodiments sequentially, in principle, the same members are denoted by the same reference numerals, and description thereof is omitted.

  Note that the signals transmitted by the speech encoding apparatus according to Embodiments 1 to 3 are essentially the same. In addition, the signals received by the speech decoding apparatuses according to the fourth and fifth embodiments are essentially the same. Therefore, as the speech encoding / decoding device according to the sixth embodiment, an arbitrary combination of the speech encoding device and the speech decoding device described above may be assumed.

(Embodiment 1)
FIG. 1 is a functional configuration diagram of the speech encoding device 111 according to the present embodiment.

  As shown in the figure, the speech encoding device 111 includes a microphone 121, an A / D conversion unit 123, a prediction analysis unit 125, a band filter unit 127, a noise determination unit 129, an encoding A switch unit 131, and the like. , A flag and gain generation unit 133, a total conversion unit 135, an encoding unit 137, and a transmission unit 139.

  The prediction analysis unit 125 includes a prediction analysis inverse filter calculator 141.

  The band filter unit 127 includes a first band filter 143, a second band filter 145, and necessary band filters after the third band filter, which are omitted in FIG.

  The noise discriminating unit 129 includes a first noise discriminator 147, a second noise discriminator 149, and a necessary noise discriminator after the third noise discriminator, although omitted in FIG.

  The encoding A switch unit 131 includes a first A switch 151, a second A switch 153, and necessary A switches after the third A switch, which are omitted in FIG.

  The flag and gain generation unit 133 includes a first flag generation and first gain calculator 155, a second flag generation and second gain calculator 157, and a third flag generation and third gain calculation, which are omitted in FIG. And a necessary flag generator and gain calculator after the counter, and further includes a flag and noise gain aggregator 159.

  The total conversion unit 135 includes a non-noise residual signal totalizer 161, a spectrum converter 163, and a non-noise band cutout unit 162.

The analog audio signal input to the microphone 121 is converted into a digital audio signal by the A / D conversion unit 123, for example, by 16 kHz sampling, and then delivered to the prediction analysis unit 125. The prediction analysis unit 125 performs prediction analysis such as linear prediction analysis or MLSA (Mel Log Spectrum Approximation) analysis on the delivered digital audio signal. The prediction analysis unit 125 divides the digital audio signal into predetermined time intervals (for example, 5 ms), and S _i = {s _{i, 0} ,..., Si _{, l−1} } (0 ≦ i ≦ M -1), a prediction coefficient such as a linear prediction coefficient or an MLSA coefficient is calculated for each time interval. Subsequently, a prediction analysis inverse filter is obtained from the prediction coefficient by the prediction analysis inverse filter calculator 141. Next, as a result of the digital audio signal S _i being input to the prediction analysis inverse filter, residual signals D _i = {d _{i, 0} ,..., D _{i, l−1} } (0 ≦ i ≦ M-1) is obtained. As described above, the prediction analysis unit 125 decomposes the digital audio signal into a prediction coefficient and a residual signal.

  The prediction coefficient generated by the prediction analysis unit 125 is transferred to the encoding unit 137 as it is.

  On the other hand, the generated residual signal is transferred to the band filter unit 127 and divided into residual signals for each band. The band filter unit 127 divides the residual signal into bands 1 to 8, for example, band 1 is 0 to 1 kHz, band 2 is 1 to 2 kHz, band 3 is 2 to 3 kHz, band 4 is 3 to 4 kHz, band 5 Is preferably 4 to 5 kHz, band 6 is 5 kHz to 6 kHz, band 7 is 6 kHz to 7 kHz, and band 8 is 7 kHz to 8 kHz. The residual signal is passed through the first band-pass filter 143 to generate a band 1 residual signal, and the residual signal is passed through the second band-pass filter 145 to generate a band 2 residual signal. The same applies hereinafter.

  A variable for band identification is denoted by ω. For example, it is assumed that the signal generated by the first band filter 143 is a signal in the band of ω = 1, and the signal generated by the second band filter 145 is a signal in the band of ω = 2.

  The generated band 1 residual signal is transferred to the first noise discriminator 147 in the noise discriminating unit 129 and the first A switch 151 in the coding A switch unit 131, and the generated band The second residual signal is transferred to the second noise discriminator 149 in the noise discriminating unit 129 and the second A switch 153 in the coding A switch unit 131, and so on.

  The first noise discriminator 147 discriminates whether or not the handed over band 1 residual signal is noise, and sends a switching command to the first A switch 151. Here, the switching command means that when it is determined that the residual signal of band 1 is noise, the first A switch 151 is closed to the a1 side, and it is determined that the residual signal of band 1 is not noise. Is a command for controlling the first A switch 151 to close the first A switch to the b1 side. The second noise discriminator 149 discriminates whether or not the handed over band 2 residual signal is noise, and sends a switching command to the second A switch 153. Here, the switching command means that when it is determined that the residual signal of band 2 is noise, the second A switch 153 is closed to the a2 side, and it is determined that the residual signal of band 2 is not noise. Is a command for controlling the second A switch 153 to close the second A switch to the b2 side. The same applies to bands 3 and after.

  When the first A switch 151 is closed to the a1 side, the band 1 residual signal is delivered to the first flag generation and first gain calculator 155 in the flag and gain generation unit 133, but the total conversion unit 135 is not transferred to the non-noise residual signal totalizer 161. On the other hand, when the first A switch 151 is closed to the b1 side, the band 1 residual signal is not delivered to the first flag generation and first gain calculator 155 in the flag and gain generation unit 133. The result is delivered to the non-noise residual signal totalizer 161 in the total conversion unit 135. When the second A switch 153 is closed to the a2 side, the band 2 residual signal is delivered to the second flag generation and second gain calculator 157 in the flag and gain generation unit 133, but the total conversion unit 135 is not transferred to the non-noise residual signal totalizer 161. On the other hand, when the second A switch 153 is closed to the b2 side, the band 1 residual signal is not delivered to the second flag generation and second gain calculator 157 in the flag and gain generation unit 133. The result is delivered to the non-noise residual signal totalizer 161 in the total conversion unit 135. The same applies to bands 3 and after.

  The first flag generator and first gain calculator 155 in the flag and gain generation unit 133 generates a flag indicating that the band 1 is a noise band when the residual signal of the band 1 is received. The gain of the residual signal of band 1 is calculated, and the flag and the gain are passed to the flag and noise gain totalizer 159 in the flag and gain generation unit 133, but the residual signal of band 1 is not received. In case you do nothing. When the second flag generation and second gain calculator 157 in the flag and gain generation unit 133 receives the residual signal of the band 2, the second flag generation and second gain calculator 157 generates a flag indicating that the band 2 is a noise band. The gain of the residual signal of band 2 is calculated, and the flag and the gain are passed to the flag and noise gain totalizer 159 in the flag and gain generation unit 133, but the residual signal of band 2 is not received. In case you do nothing. The same applies to bands 3 and after.

  The gain of each band represents the intensity of the component of the band of the residual signal. In the case of an audio signal, generally, the gain becomes a different value if the band is different. The gain is transmitted to a speech decoding device described later. Then, the apparatus reproduces an audio signal reflecting the difference in intensity of each band of the original residual signal. Therefore, obtaining the gain for each band by the speech encoding device 111 means that the speech decoding device, which will be described later, has a higher quality speech signal than when assuming that the gain is a constant value independent of the bandwidth. Contribute to playing. In the present embodiment, the characteristics of the residual signal in the non-noise band are transmitted to the speech decoding apparatus as frequency conversion coefficients represented by numerical values, so that it is not necessary to obtain a separate gain.

  Since the noise discriminating unit 129 and the encoding A switch unit 131 operate as described above, the flag and noise gain counter 159 collects flags and gains in the noise band. The flags and gains in these noise bands are delivered to the encoding unit 137.

  The non-noise residual signal totalizer 161 in the totalizing conversion unit 135 grasps which band residual signal has been delivered from the encoding A switch unit 131, and transmits only the residual signal in the band in real time. A signal superimposed on the region is generated.

  Since the noise discriminating unit 129 and the encoding A switch unit 131 operate as described above, the band recognized by the non-noise residual signal totalizer 161 as described above is a non-noise band as a result. Similarly, the signal generated by superimposing the non-noise residual signal totalizer 161 as described above is a residual signal in the non-noise band as a result.

  The residual signal in the non-noise band is transferred from the non-noise residual signal totalizer 161 to the spectrum converter 163 in the total conversion unit 135, and is subjected to fast Fourier transform (FFT), discrete cosine transform (DCT). , Discrete Cosine Transform), modified cosine transform (MDCT), and the like, and after being converted into frequency conversion coefficients, it is delivered to the non-noise band cutout unit 162 in the total conversion unit 135. .

  The non-noise band cutout unit 162 is also handed over information on which band is the non-noise band, as grasped by the non-noise residual signal totalizer 161 as described above. The non-noise band cutout unit 162 refers to such information, extracts only the frequency conversion coefficient in the non-noise band from the frequency conversion coefficients transferred from the spectrum converter 163, and transfers the extracted frequency conversion coefficient to the encoding unit 137. That is, only the frequency conversion coefficient in the non-noise band is extracted and delivered to the encoding unit 137 out of the frequency conversion coefficients in the entire band obtained by frequency-converting the residual signal in the non-noise band.

  However, since the noise determination unit 129 and the encoding A switch unit 131 operate as described above, the residual signal in the non-noise band generated by the non-noise residual signal totalizer 161 by the above-described superposition is Originally, no noise band component is included. Therefore, in principle, the frequency conversion coefficients generated by the spectrum converter 163 are those in which numerical values 0 are arranged in the noise band. Therefore, the non-noise band cutout unit 162 does not receive the information on the non-noise band from the non-noise residual signal totalizer 161, and calculates a coefficient having a value of 0 from the frequency conversion coefficient transferred from the spectrum converter 163. What has been removed may be delivered to the encoding unit 137. However, in consideration of the accuracy of the band filter unit 127 as a band filter, in the present embodiment, in order to ensure the operation, the non-noise band cutout unit 162 relates to the non-noise band in cutting out the frequency conversion coefficient. Information shall be referenced.

  The encoding unit 137 receives the prediction coefficient from the prediction analysis unit 125, the flag and gain in the noise band from the flag and gain generation unit 133, and the frequency conversion coefficient in the non-noise band from the total conversion unit 135. The encoding unit 137 collectively encodes them using a predetermined encoding method, for example, a vector quantization method or a Huffman encoding method, and delivers the generated code to the transmission unit 139.

  The transmission unit 139 receives, from the encoding unit 137, a prediction coefficient, a flag, a gain, and a frequency conversion coefficient in a non-noise band, which are encoded, and relates to this according to Embodiment 4 or 5 described later. Transmission is performed toward the speech decoding apparatus or another speech encoding / decoding apparatus according to Embodiment 6 that functions as the speech decoding apparatus. In this embodiment, the transmission method is wireless communication. However, various other communication methods such as wired or a combination of wired and wireless may be used.

  The speech encoding apparatus 111 according to the present embodiment divides the residual signal into a plurality of bands, and for the noise band, only the flag indicating that and the gain in the band are to be transmitted, and the frequency for the non-noise band By using the transform coefficient as a transmission target, it is possible to achieve both efficient transmission of information related to the residual signal and ensuring the quality of reproduced audio.

(Embodiment 2)
FIG. 2 is a functional configuration diagram of the speech encoding device 113 according to the present embodiment. The speech coding apparatus 113 generally has the same configuration as the speech coding apparatus 111 according to the first embodiment. The main difference is that the encoding A switch 131 is replaced by an encoding C switch unit 165, and that a constant voltage source (voltage V _C ) is provided accordingly, and that the Band Elimination Filter 169 is provided. This is that the specifications of the constituent elements have been partially changed after the parts that have been collected as the total conversion unit 135 have been disassembled.

  In the first embodiment, the noise determination unit 129 sends a switching command to the encoding A switch unit 131. However, in this embodiment, the noise determination unit 129 sends a selection ON command to the encoding C switch unit 165. send. Here, if the selection ON command is for the band 1, when it is determined that the residual signal of the band 1 is noise, the c1 of the first C switch 171 is closed and the d1 is opened. This is a command for controlling the first C switch 171 to open c1 of the first C switch 171 and close d1 when it is determined that the residual signal is not noise. In the case of band 2, if it is determined that the residual signal of band 2 is noise, c2 of second C switch 173 is closed and d2 is opened, and the residual signal of band 2 is not noise. Is determined to control the second C switch 173 to open c2 of the second C switch 173 and close d2. The same applies to the band 3 and subsequent bands.

When c1 is closed and d1 is opened in the first C switch 171, the residual signal of band 1 is delivered to the flag and gain generation unit 133, but the voltage V _C is not applied to the non-noise band determiner 167, The noise band determiner 167 stores that band 1 is a noise band. On the other hand, when c1 is opened and d1 is closed in the first C switch 171, the residual signal of band 1 is not delivered to the flag and gain generation unit 133, and the voltage V _C is applied to the non-noise band determiner 167. The noise band determiner 167 stores that band 1 is a non-noise band. When c2 is closed and d2 is opened in the second C switch 173, the residual signal of band 2 is delivered to the flag and gain generation unit 133, but the voltage V _C is not applied to the non-noise band determiner 167, and non-noise is determined. The noise band determiner 167 stores that band 2 is a noise band. On the other hand, when c2 is opened and d2 is closed in the second C switch 173, the residual signal of the band 2 is not delivered to the flag and gain generation unit 133, and the voltage V _C is applied to the non-noise band determiner 167. The noise band determiner 167 stores that band 2 is a non-noise band. The same applies to bands 3 and after.

Thus, the constant voltage source (voltage V _C ) is used by the non-noise band determiner 167 to grasp which band is the non-noise band. The illustrated constant voltage source is schematic and may be replaced by another mechanism as long as the non-noise band determiner 167 enables such grasping.

  The non-noise band determiner 167 delivers information regarding the recognized non-noise band to the non-noise band cutout unit 162 in the same manner as the non-noise residual signal totalizer 161 performs in the first embodiment.

  However, unlike the non-noise residual signal aggregator 161 in the first embodiment, the non-noise band determiner 167 does not receive a non-noise band residual signal. In the present embodiment, in order for the spectrum converter 163 to acquire a residual signal in a non-noise band, first, the entire residual signal is delivered from the prediction analysis unit 125 to the Band Elimination Filter 169. Since the non-noise band determiner 167 knows which band is the non-noise band, it can be said that, on the contrary, it knows which band is the noise band. Therefore, the non-noise band determiner 167 sends a command for collectively specifying such a noise band, that is, a band to be deleted before the residual signal is input to the spectrum converter 163, to the Band Elimination Filter 169. The Band Elimination Filter 169 is a filter that can freely select and set a band to be deleted. In the present embodiment, the band Elimination Filter 169 is a band to be deleted from the residual signal delivered from the prediction analysis unit 125 according to the command. And is transferred to the spectrum converter 163.

  This embodiment can be said to be a modification of the first embodiment. However, in the present embodiment, unlike the case of the first embodiment, the non-noise residual signal input to the spectrum converter 163 is not passed through the band filter unit 127 but is passed through the Band Elimination Filter 169. Something is used. Therefore, when only one Band Elimination Filter having higher performance as a filter than the many band filters used in the band filter unit 127 can be obtained, this embodiment is more suitable in the case of the first embodiment. There is an advantage that the frequency conversion coefficient in the non-noise band can be obtained with high accuracy as compared with. In the case of the first embodiment, an error may occur when the non-noise residual signal totalizer 161 performs the superimposing process of the residual signal in the non-noise band, but in the case of the present embodiment, the superimposing process is performed. In this respect, it is expected that the frequency conversion coefficient can be obtained with higher accuracy.

(Embodiment 3)
FIG. 3 is a functional configuration diagram of the speech encoding device 115 according to the present embodiment. The speech coding apparatus 115 has a configuration similar to that of the speech coding apparatus 111 according to the first embodiment, except that the coding A switch 131 is replaced with a coding E switch unit 177, and an aggregation conversion unit. It is different in that the specifications of the constituent elements excluding the spectrum converter 163 are partly changed after the part grouped as 135 is completely decomposed.

  In the present embodiment, the residual signal from the prediction analysis unit 125 is directly delivered to the spectrum converter 163. The spectral converter 163 differs from the previous two embodiments in that the frequency conversion coefficient is calculated from the entire residual signal without regard to whether the handed over residual signal contains a noise band or not. Ask for.

  The obtained frequency conversion coefficient is delivered to the frequency conversion coefficient discriminator 175. The frequency conversion coefficient discriminator 175 is similar to the non-noise band cutout unit 162 in the previous two embodiments in that the received frequency conversion coefficient is associated with a predetermined band. However, while the latter deletes the frequency conversion coefficients belonging to the non-noise band, the former only classifies all the frequency conversion coefficients into a plurality of predetermined bands, and the frequency conversion coefficient in band 1 and band 2 The difference is that all frequency conversion coefficients are output in the end, as in FIG.

  The signal line for transmitting the frequency conversion coefficient of band 1 output from the frequency conversion coefficient discriminator 175 is connected to the first E switch 181 in the encoding E switch unit 177, and the frequency conversion coefficient of band 2 is transmitted. The signal line to be connected is connected to the second E switch 183. The same applies to bands 3 and after.

  In the first embodiment, the noise determination unit 129 sends a switching command to the encoding A switch unit 131. However, in this embodiment, the noise determination unit 129 sends a selection on command to the encoding E switch unit 177. send. Here, if the selection ON command is for the band 1, if it is determined that the residual signal of the band 1 is noise, the e1 of the first E switch 181 is closed and the f1 is opened. This is a command for controlling the first E switch 181 to open e1 of the first E switch 181 and close f1 when it is determined that the residual signal is not noise. In the case of band 2, if it is determined that the residual signal of band 2 is noise, e2 of the second E switch 183 is closed and f2 is opened, and the residual signal of band 2 is not noise. Is determined to control the second E switch 183 to open e2 of the second E switch 183 and close f2. The same applies to the band 3 and subsequent bands.

  When e1 is closed and f1 is opened in the first E switch 181, the residual signal of band 1 is delivered to the flag and gain generation unit 133, but the non-noise frequency conversion coefficient totalizer 179 stores the frequency conversion coefficient in band 1. Not handed over, the non-noise frequency conversion coefficient totalizer 179 does not store the frequency conversion coefficients in band 1. On the other hand, when e1 is opened and f1 is closed in the first E switch 181, the residual signal of band 1 is not delivered to the flag and gain generation unit 133, and the frequency conversion coefficient in band 1 is not transferred to the non-noise frequency conversion coefficient totalizer 179. Is passed, and the non-noise frequency conversion coefficient totalizer 179 stores the frequency conversion coefficients in band 1. When e2 is closed and f2 is opened in the second E switch 183, the residual signal in band 2 is delivered to the flag and gain generation unit 133, but the frequency conversion coefficient of band 2 is stored in the non-noise frequency conversion coefficient totalizer 179. Not handed over, the non-noise frequency conversion coefficient totalizer 179 does not store the frequency conversion coefficients in band 2. On the other hand, when e2 is opened and f2 is closed in the second E switch 183, the residual signal of the band 2 is not delivered to the flag and gain generation unit 133, and the frequency conversion coefficient in the band 2 is input to the non-noise frequency conversion coefficient totalizer 179. Is passed, and the non-noise frequency conversion coefficient totalizer 179 stores the frequency conversion coefficients in band 2. The same applies to bands 3 and after.

  As a result, the non-noise frequency conversion coefficient totalizer 179 counts and stores frequency conversion coefficients in the non-noise band. The non-noise frequency conversion coefficient totalizer 179 passes the frequency conversion coefficient to the encoding unit 137.

  What the speech encoding apparatus 115 according to the present embodiment encodes is the same as what the speech encoding apparatus 111 and the speech encoding apparatus 113 according to the previous two embodiments encode. is there.

  However, in the case of the present embodiment, the frequency conversion coefficient is obtained collectively over the entire band before the residual signal is divided into the band-specific residual signals. Therefore, unlike the case of the first embodiment, the frequency conversion coefficient can be obtained with high accuracy regardless of the performance of the band filter unit 127. Further, in the present embodiment, the frequency conversion coefficient is obtained with high accuracy without being affected by errors due to signal superposition processing in the non-noise residual signal totalizer 161 (FIG. 1) in the first embodiment. . Furthermore, even in comparison with the second embodiment, in this embodiment, the frequency conversion coefficient can be obtained with high accuracy irrespective of the performance of the band elimination filter.

(Embodiment 4)
FIG. 4 is a functional configuration diagram of the speech decoding apparatus 211 according to the present embodiment.

  As shown in the figure, the speech decoding apparatus 211 includes a receiving unit 221, a decoding unit 223, a flag presence / absence determination unit 225, a decoding G switch unit 227, a band-specific noise sequence generation unit 229, and an aggregate inverse conversion unit 231. A residual signal restoration unit 233, a synthesis filter calculation unit 235, a synthesis filter unit 237, a D / A conversion unit 239, and a speaker 241.

  The flag presence / absence determining unit 225 includes a first flag presence / absence determiner 243, a second flag presence / absence determiner 245, and a necessary flag presence / absence determiner after the third flag presence / absence determiner, which are omitted in FIG. .

  The decoding G switch unit 227 includes a first G switch 247, a second G switch 249, and a necessary G switch after the third G switch, although omitted in FIG.

  The band-specific noise sequence generation unit 229 includes a first noise sequence generator 251, a second noise sequence generator 253, a noise sequence generator required after the third noise sequence generator, although omitted in FIG. Is provided.

  The tabulation inverse conversion unit 231 includes a frequency conversion coefficient tabulation and supplementer 255 and a spectrum inverse converter 257.

  The receiving unit 221 includes the speech encoding device 111 (FIG. 1) according to the first embodiment, the speech encoding device 113 (FIG. 2) according to the second embodiment, and the speech encoding device 115 (FIG. 3) according to the third embodiment. However, the code transmitted from the transmission unit 139 by wireless communication means or the like is received and delivered to the decoding unit 223. Such a code is obtained by encoding a prediction coefficient, a flag, a gain, and a frequency conversion coefficient in a non-noise band.

  The decoding unit 223 decodes the code delivered from the reception unit 221 and, as described above, the prediction coefficient, flag, gain, and gain in each time segment identified by the variable i (0 ≦ i ≦ M−1). And a frequency conversion coefficient in a non-noise band.

  The decoding unit 223 delivers the generated prediction coefficient to the synthesis filter calculation unit 235. At the same time, the decoding unit 223 delivers the generated flag, gain, and frequency conversion coefficient in the non-noise band to the flag presence / absence determination unit 225 and the decoding G switch unit 227 as information for each band. In general, information regarding the presence / absence of a flag among the information of each band is delivered to the flag presence / absence determination unit 225, and information relating to other than the flag among the information of each band is delivered to the decoding G switch unit 227.

  Note that the flag, gain, and frequency conversion coefficient may or may not be generated by the decoding unit 223 in each band. That is, in the transmission side device such as the speech encoding device 111 (FIG. 1), the flag and the gain are generated and encoded only for the noise band, and the frequency conversion coefficient is generated and encoded only for the non-noise band. Therefore, even if decoding by the decoding unit 223 is performed in the speech decoding apparatus 211 according to the present embodiment which is a receiving-side apparatus, the frequency conversion coefficient in the noise band and the flag and gain in the non-noise band are generated. It will never be done.

  One of the roles of the decoding unit 223 is to notify the first flag presence / absence discriminator 243 in the flag presence / absence discriminating unit 225 of the presence / absence of the flag in the band 1 in the band 1 information. More precisely, the decoding unit 223 notifies the first flag presence / absence discriminator 243 when the flag in the band 1 is generated, and the first flag when the flag in the band 1 is not generated. No notification is made to the presence / absence discriminator 243. For the band 2, the decoding unit 223 notifies the second flag presence / absence discriminator 245 when the flag for the band 2 is generated, and the second flag when the flag for the band 2 is not generated. No notification is made to the presence / absence discriminator 245. The same applies to bands 3 and after.

  The decoding unit 223 also includes information other than the flag in the band 1 information, that is, the gain in the band 1 when the band 1 is a noise band, and the band 1 when the band 1 is a non-noise band. Is notified to the first G switch 247 in the decoding G switch unit 227. For band 2, information other than the flag, that is, gain in band 2 when band 2 is a noise band, and frequency conversion coefficient in band 2 when band 2 is a non-noise band are decoded. The second G switch 249 in the G switch unit 227 is notified. The same applies to bands 3 and after.

  The first flag presence / absence discriminator 243 determines whether or not a notification indicating that the band 1 flag has been generated is received from the decoding unit 223, and sends a switching command to the first G switch 151. Here, when it is determined that the notification that the band 1 flag has been generated is received, the switching command closes the first G switch 247 to the g1 side, and notifies that the band 1 flag has been generated. When it is determined that the first G switch is not received, the first G switch 247 is controlled to close the first G switch to the h1 side. The second flag presence / absence determiner 245 determines whether or not a notification that the band 2 flag has been generated is received from the decoding unit 223, and sends a switching command to the second G switch 249. Here, when it is determined that the notification that the band 2 flag has been generated is received, the switching command closes the second G switch 249 to the g2 side, and notifies that the band 2 flag has been generated. The command is for controlling the second G switch 249 so that the second G switch is closed to the h2 side when it is determined that it has not been received. The same applies to bands 3 and after.

  When the first G switch 247 is closed to the g1 side, the gain of the band 1 is delivered to the first noise string generator 251 in the noise string generator 229 for each band. This is because the first G switch 247 is closed to the g1 side because the first flag presence / absence discriminator 243 detects the presence of the flag in the band 1 as described above. That is, the unit 223 decodes the gain instead of the frequency conversion coefficient in the band 1, and therefore, it is not the frequency conversion coefficient but the gain that is notified to the first G switch 247 as the band 1 information from the decoding unit 223. Therefore, since the first G switch 247 is closed to the g1 side, the information flowing through the signal line connected between the decoding unit 223 and the first noise string generator 251 is the gain of band 1. It is.

  On the other hand, when the first G switch 247 is closed to the h1 side, the frequency conversion coefficient in the band 1 is delivered to the frequency conversion coefficient totalization and supplementer 255 in the totalization inverse conversion unit 231. This is because the first G switch 247 is closed to the h1 side because the first flag presence / absence discriminator 243 detects the absence of the flag in the band 1 as described above. That is, the unit 223 decodes the frequency conversion coefficient instead of the gain in the band 1, and therefore, it is not the gain but the frequency conversion coefficient that is notified to the first G switch 247 as the band 1 information from the decoding unit 223. Therefore, the information flowing through the signal line connected between the decoding unit 223 and the frequency conversion coefficient totaling and supplementing device 255 when the first G switch 247 is closed to the h1 side is a frequency conversion coefficient in the band 1. Because it becomes.

  Similarly, when the second G switch 249 is closed to the g2 side, the gain of the band 2 is delivered to the second noise string generator 253 in the noise string generator 229 for each band. This is because the second G switch 249 is closed to the g2 side because the second flag presence / absence discriminator 245 detects the presence of the flag in the band 2 as described above. That is, the unit 223 decodes the gain, not the frequency conversion coefficient, in the band 2, and therefore it is not the frequency conversion coefficient but the gain that is notified from the decoding unit 223 to the second G switch 249 as the band 2 information. Therefore, since the second G switch 249 is closed to the g2 side, information flowing through the signal line connected between the decoding unit 223 and the second noise string generator 253 is a gain of band 2. It is.

  On the other hand, when the second G switch 249 is closed to the h2 side, the frequency conversion coefficient in the band 2 is delivered to the frequency conversion coefficient totalization and supplementer 255 in the totalization inverse conversion unit 231. This is because the second G switch 249 is closed to the h2 side because the second flag presence / absence discriminator 245 detects the absence of the flag in the band 2 as described above. The unit 223 decodes the frequency conversion coefficient instead of the gain in the band 2. Therefore, it is not the gain but the frequency conversion coefficient that is notified to the first G switch as the band 2 information from the decoding unit 223. Therefore, when the second G switch 249 is closed to the h2 side, information flowing through the signal line connected between the decoding unit 223 and the frequency conversion coefficient totaling and supplementing device 255 is a frequency conversion coefficient in the band 2. Because it becomes.

  The same applies to bands 3 and after.

  When receiving the gain of band 1, the first noise string generator 251 in the noise string generation unit 229 for each band receives the built-in white noise generator (not shown) and the band 1 from the white noise. A built-in band-pass filter (not shown) for extracting a signal and a built-in multiplier (not shown) that adjusts the amplitude of the signal according to the received gain generate a noise sequence of band 1; On the other hand, if the gain of the band 1 is not received while performing the transfer to the residual signal restoration unit 233, nothing is performed.

  When the second noise sequence generator 253 in the noise sequence generation unit 229 for each band receives the gain of the band 2, the second noise string generator 253 includes a built-in white noise generator (not shown) and the band 2 from the white noise. A built-in band-pass filter (not shown) for extracting the signal and a built-in multiplier (not shown) that adjusts the amplitude of the signal according to the received gain generate a noise sequence of band 2; On the other hand, if the gain of band 2 is not received, nothing is performed.

  The same applies to bands 3 and after.

  Since the flag presence / absence determination unit 225 and the decoding G switch unit 227 operate as described above, the residual signal restoration unit 233 receives a signal that is considered valid as a residual signal in the noise band in all noise bands. It will be input across. The expression “valid” is used here for the following reason. That is, in the transmission-side apparatus such as the speech encoding apparatus 111 (FIG. 1) according to the first embodiment, the residual signal in the noise band is characterized only by the amount of gain, and only this gain is the reception-side apparatus. This is transmitted to the speech decoding apparatus 211 according to the embodiment. Therefore, the speech decoding apparatus 211 cannot completely restore the residual signal in the noise band, and in a situation where only the amount of gain is notified, the most appropriate residual signal, that is, a signal that is considered appropriate, Alternatively, it can be said that a pseudo residual signal is generated.

  The frequency conversion coefficient totaling and replenisher 255 in the totalizing inverse conversion unit 231 stores the frequency conversion coefficient in the band 1 when it is received, and stores the frequency conversion coefficient in the band 1 when it is not received. The frequency conversion coefficients in band 1 are all set to 0 and stored. The frequency conversion coefficient totaling and replenisher 255 in the totalizing inverse conversion unit 231 stores the frequency conversion coefficient in the band 2 when it is received, and stores the frequency conversion coefficient in the band 2 when it is not received. The frequency conversion coefficients in band 2 are all set to 0 and stored. The same applies to bands 3 and after.

  In this way, the frequency conversion coefficient totaling and supplementing unit 255 stores the frequency conversion coefficient as it is for the band that has received the frequency conversion coefficient, and sets 0 as the frequency conversion coefficient for the band that has not received the frequency conversion coefficient. By supplementing and storing, a frequency conversion coefficient without a missing band is generated. The frequency conversion coefficient totalizing and supplementing unit 255 delivers the frequency conversion coefficient to the spectral inverse converter 257 that is also in the totaling inverse conversion unit 231. The spectrum inverse converter 257 is an inverse conversion technique that is paired with a predetermined frequency conversion technique used in the spectrum converter 163 in the transmission-side apparatus such as the speech encoding apparatus 111 (FIG. 1) according to the first embodiment. Is used to inversely transform the passed frequency transform coefficient into a real time domain signal. Since the frequency conversion coefficient received by the spectrum converter 163 is not missing as described above, the inverse conversion is performed smoothly. The spectrum inverse transformer 257 delivers the real-time domain signal to the residual signal restoration unit 233.

  Since the flag presence / absence determination unit 225 and the decoding G switch unit 227 operate as described above, the signal in the real time domain transferred from the spectrum inverse converter 257 to the residual signal restoration unit 233 is restored in the non-noise band. It is a residual signal.

From the above, the residual signal in the noise band is handed over to the residual signal restoring unit 233 from the noise sequence generating unit 229 for each band although it is a pseudo signal as described above, and the non-noise is output from the total inverse converting unit 231. Since the restored residual signal in the band is delivered, the residual signal in the entire band is eventually delivered. The residual signal restoration unit 233 superimposes the residual signals of these bands, thereby restoring the residual signal D ′ _i = {d ′ _{i, 0} ,..., D ′ _{i, l−1} } (0 ≦ i ≦ M−1) is generated. The generated restored residual signal is delivered to the synthesis filter unit 237.

  The prediction coefficient is delivered from the decoding unit 223 to the synthesis filter calculation unit 235. The synthesis filter calculation unit 235 determines the specification of the synthesis filter by any known method based on the delivered prediction coefficient, and notifies the synthesis filter unit 237 of the result of the determination. The synthesizing filter unit 237 determines its own specification according to the notification.

  The restoration residual signal from the residual signal restoration unit 233 is input to the synthesis filter unit 237 as an excitation signal. As already described, the residual signal and the excitation signal are merely the same signal viewed from different viewpoints. As a result of inputting the excitation signal, the synthesis filter unit 237 generates a restored digital audio signal. The signal is converted into an analog audio signal by the D / A converter 239 and then sent to the speaker 241. Thus, the restored audio signal is emitted from the speaker 241 in a manner that can be heard by the human ear.

(Embodiment 5)
FIG. 5 is a functional configuration diagram of the speech decoding apparatus 213 according to the present embodiment.

  This embodiment is a modification of the fourth embodiment. That is, in the fourth embodiment, a real-time signal is generated for each noise band. In the present embodiment, a frequency conversion coefficient is generated in accordance with the gain of the noise band, and then the non-noise band is generated. Reverse conversion to the real time domain is performed together with the frequency conversion coefficient.

  Compared with the speech decoding apparatus 211 (FIG. 4) according to the fourth embodiment, the speech decoding apparatus 213 according to the present embodiment replaces the band-specific noise sequence generation unit 229 (FIG. 4) with the constant frequency conversion coefficient generation unit 259 for each band. Replace, partially change the configuration of the signal line downstream of the decoding G switch unit 227, replace the total inverse conversion unit 231 (FIG. 4) with the collective total inverse conversion unit 261, and omit the residual signal restoration unit 233 It has been configured.

  When the first constant frequency conversion coefficient generator 263 in the band-specific constant frequency conversion coefficient generator 259 receives the gain of the band 1, first, the unit amount of the frequency conversion coefficient is arranged as the frequency conversion coefficient in the band 1. And then multiplying the gain to generate a frequency conversion coefficient in band 1 and hand it over to the frequency conversion coefficient totalizer 267 in the batch total inverse transform unit 261 while receiving the gain in band 1 If not, no action is taken.

  When the second constant frequency conversion coefficient generator 265 in the band-specific constant frequency conversion coefficient generation unit 259 receives the gain of the band 2, first, the unit amount of the frequency conversion coefficient is arranged as the frequency conversion coefficient in the band 2. And then multiplying the gain to generate a frequency conversion coefficient in band 2 and hand it over to the frequency conversion coefficient totalizer 267 in the batch total inverse transform unit 261 while receiving the gain in band 2 If not, no action is taken.

  The same applies to bands 3 and after.

  Since the signal line connecting the decoding G switch unit 227 and the batch aggregation inverse conversion unit 261 is provided with a connection unit indicated by g′1 or g′2 in FIG. Is possible. In the first G switch 247, the switch is closed only on one side of g1 and h1, and in the second G switch 249, the switch is closed only on either side of g2 or h2. Only one of the signal from the decoding G switch unit 227 and the signal from the constant frequency conversion coefficient generation unit 259 for each band flows in, and the two types of signals are not mixed.

  The frequency conversion coefficient totalizer 267 in the batch total inverse conversion unit 261 is very similar to the frequency conversion coefficient totalization and supplementer 255 in the total inverse conversion unit 231 in the speech decoding apparatus 211 (FIG. 4) according to the fourth embodiment. It has a function. The frequency conversion coefficient tabulation and supplementer 255 (FIG. 4) in the fourth embodiment needs to generate a frequency conversion coefficient composed of a numerical value 0 in order to compensate for the missing band.

  However, in the case of the present embodiment, the frequency conversion coefficient totalizer 267 has received, as is apparent from the operations of the flag presence / absence determination unit 225, the decoding G switch unit 227, and the constant frequency conversion coefficient generation unit 259 for each band. By simply storing the transform coefficient, a frequency transform coefficient without a missing band can be generated and used for inverse transform by the spectrum inverse transformer 257.

  As is apparent from the operations of the flag presence / absence determination unit 225, the decoding G switch unit 227, and the constant frequency conversion coefficient generation unit 259 for each band, the spectrum inverse converter 257 in the collective aggregation inverse conversion unit 261 generates The residual signal already contains both noise band and non-noise band components.

  Therefore, the residual signal may be delivered to the synthesis filter unit 237 as a restored residual signal or an excitation signal. In the case of the present embodiment, the residual signal restoration unit 233 (FIG. 4) used in Embodiment 4 to generate the restored residual signal by superimposing the noise band component and the non-noise band component is unnecessary. It is.

  Thus, the present embodiment performs the same operation as that of the fourth embodiment in principle. However, in the fourth embodiment, a white noise generation source and a band filter (both not shown) are necessary in the band-specific noise string generation unit 229 (FIG. 4). Since such a member is unnecessary, it is simpler.

(Embodiment 6)
FIG. 6 shows a speech encoding / decoding device 311 according to this embodiment. Speech encoding apparatus 111, speech encoding apparatus 113, speech encoding apparatus 115, speech decoding apparatus 211 according to Embodiments 1 to 5 that have been described with reference to FIGS. The speech decoding device 213 is physically realized by the speech encoding / decoding device 311 according to the present embodiment, which integrates the functions of both types of devices from the viewpoint of usability. In the following description, a mobile phone is assumed as the speech encoding / decoding device 311.

  The speech encoding / decoding device 311 includes a microphone 121 already shown in FIGS. 1 to 3 and a speaker 241 already shown in FIGS. 4 and 5. The apparatus further includes an antenna 335 and operation keys 337. The apparatus includes a CPU 321, a ROM (Read Only Memory) 323, a storage unit 325, a voice processing unit 329, a wireless communication unit 327, an operation key input processing unit 331, which are connected to each other via a system bus 333. Are further provided. The storage unit 325 includes, for example, a RAM (Random Access Memory) 339 and a hard disk 341. The speech encoding / decoding device 311 may include, for example, the Band Elimination Filter 169 shown in Embodiment 2 as dedicated hardware separately from the one shown in FIG.

  The ROM 323 stores an operation program for voice encoding and decoding. The CPU 321 operates according to the operation program. Then, the CPU 321 performs numerical computation while appropriately exchanging data between a built-in register (not shown) and the storage unit 325, and the speech encoding / decoding device 311 is shown in FIGS. 1 to 5. The functions of the speech encoding device 111, the speech encoding device 113, the speech encoding device 115, the speech decoding device 211, and the speech decoding device 213 are exhibited. At that time, the CPU 321 exchanges data with the voice processing unit 329, the wireless communication unit 327, and the operation key input processing unit 331 as necessary.

  The audio processing unit 329 in FIG. 6 can operate as the A / D conversion unit 123 in FIGS. 1 to 3 and the D / A conversion unit 239 in FIGS. 4 and 5. The wireless communication unit 327 can operate as the transmission unit 139 in FIGS. 1 to 3 and the reception unit 221 in FIGS. 4 and 5. The transmission / reception of the code is basically performed by wireless communication using the antenna 335 of FIG. 6, but may be performed by another method, for example, wired communication. The operation key input processing unit 331 receives an operation signal from the operation key 337 and transmits a key code signal corresponding to the operation signal to the CPU 321. The operation key 337 is used to specify a voice encoding / decoding device 311 as a communication partner, that is, to input a so-called telephone number, and basically, various set items can be set according to the user's preference. It may be used to change.

(Predictive analysis procedure)
Hereinafter, the prediction analysis performed by the prediction analysis unit 125 of FIGS. 1 to 3 will be described with reference to the flowchart shown in FIG. As prediction analysis, for example, linear prediction analysis and MLSA (Mel Log Spectrum Approximation) analysis are known. In FIG. 7, both analyzes are shown together with the latter in parentheses.

The storage unit 325 (FIG. 6) already stores digital audio signals (input waveforms) S _i = {s _{i, 0} ,..., S _{i, l−1} } (0 ≦ i ≦ M−1). Suppose that The CPU 321 (FIG. 6) uses a built-in counter register (not shown) for storing the input signal sample counter i, and sets i = 0 as an initial value (step S411 in FIG. 7).

The CPU 321 loads the input signal samples S _i = {s _{i, 0} ,..., S _{i, l−1} } from the storage unit 325 to a built-in general-purpose register (not shown) (step of FIG. 7). S413).

In the case of linear prediction analysis, the CPU 321 calculates linear prediction coefficients A _i = {a _{i, 1} ,..., A _{i, n} } from the input signal sample S _i (step S415). Where n is the order of linear predictive analysis. As a calculation method, any known method may be employed as long as the residual signal is evaluated to be sufficiently small based on a predetermined scale. For example, it is preferable to use a well-known calculation method that combines the calculation of the autocorrelation function and the Levinson-Durbin algorithm.

In the case of MLSA analysis, the CPU 321 first calculates a cepstrum C _i = {c _{i, 0} ,..., C _{i, (l / 2) −1} } from the input signal sample S _i . Any known method may be employed for such calculation. In any method, procedures such as discrete Fourier transform, absolute value, logarithm, and inverse discrete Fourier transform are generally performed. Next, MLSA filter coefficients M _i = {mi _{, 0} ,..., Mi _{, p−1} } are calculated from the obtained cepstrum C _i by any known method (step S415).

In the case of linear prediction analysis, linear prediction coefficient A _i = {a _{i, 1} ,..., A _{i, n} }, and in the case of MLSA analysis, MLSA filter coefficient M _i = {m _{i, 0} ,. m _{i, p-1} } is stored as a prediction coefficient in the storage unit 325 (step S417).

Subsequently, in the case of linear prediction analysis, an inverse prediction filter AIA _i for prediction analysis is calculated from the linear prediction coefficient A _i by an arbitrary known method. In the case of MLSA analysis, an arbitrary linear prediction filter A _i is calculated from the MLSA filter coefficient M _i . An inverse MLSA filter AIM _i for prediction analysis is calculated by a known method (step S419). These calculations correspond to the calculations performed by the prediction analysis inverse filter calculator 141 in FIGS.

By passing the input signal samples S _i = {s _{i, 0} ,..., S _{i, l-1} } through the obtained prediction analysis inverse linear prediction filter AIA _i or prediction analysis inverse MLSA filter AIM _i , Residual signal D _i = {d _{i, 0} ,..., D _{i, l−1} } is obtained (step S421 in FIG. 7). The residual signal D _i is stored in the storage unit 325 (step S423).

  Here, it is determined whether or not the input signal sample counter i has reached M−1 (step S425). If it has been reached (step S425; Yes), the process ends. On the other hand, if not reached (step S425; No), i is incremented by 1 (step S427) in order to perform processing on the input signal sample in the next time interval, and the processing after step S413 is repeated.

(Procedure for generating flags etc. from residual signal)
In the following, a flag, a gain, and a frequency conversion coefficient are generated from the residual signal, which is performed by the speech encoding device 111, the speech encoding device 113, and the speech encoding device 115 according to Embodiments 1 to 3. Explain the procedure.

As a premise, the digital audio signal (input signal) S _i = {s _{i, 0} ,..., Si _{, l−1} } (0 ≦ i) by the A / D converter 123 (FIGS. 1 to 3). ≦ M−1) is already obtained and stored in the storage unit 325 (FIG. 6), and the prediction analysis unit 125 (FIGS. 1 to 3) that has received the signal has already predicted the coefficient and the residual signal. D _i = {d _{i, 0} ,..., D _{i, l-1} }} (0 ≦ i ≦ M−1) is also obtained, and these coefficients and residual signals are also stored in the storage unit 325. It shall be.

  First, the procedure of the process performed by the speech encoding apparatus 111 (FIG. 1) according to the first embodiment will be described with reference to the flowcharts shown in FIGS.

  The CPU 321 (FIG. 6) sets an input signal sample counter i to i = 0 in a built-in counter register (not shown) (step S431 in FIG. 8).

The CPU 321 loads the residual signal D _i = {d _{i, 0} ,..., D _{i, l−1} } from the storage unit 325 (FIG. 6) to a built-in general-purpose register (not shown). (Step S433 in FIG. 8).

  In the counter register, the CPU 321 sets the band identification variable ω to ω = 1 (step S435).

The CPU 321 functions as the band filter unit 127 (FIG. 1) to generate the residual signal D _{i, ω} = {d _{i, ω, 0} ,..., D _{i, ω, l−1} } of the band ω. It is generated (step S437 in FIG. 8).

The CPU 321 functions as the noise determination unit 129 (FIG. 1) to determine whether Di _{and ω} are noise (step S439). If it is determined that D _{i and ω} are noises (step S439; Yes), the CPU 321 functions as the flag and gain generation unit 133 (FIG. 1) to generate the flags Flag _{i and ω} in the band ω. , and performs the gain G _i in the band _omega, the calculation of the _omega, the (step S441 in FIG. 8).

After the Flag _{i, ω} and the gains G _{i, ω} are stored in the storage unit (step S443), the process proceeds to step S447. When it is determined that D _{i and ω} are not noise (step S439; No), the CPU 321 functions as the total conversion unit 135 (FIG. 1) to store D _{i and ω} in the storage unit 325. (Step S445 in FIG. 8), the process proceeds to Step S447.

Various methods for determining whether Di _{and ω} are noises performed in step S439 can be considered, and a preferable example thereof is as follows. That is, the normalized autocorrelation function for the residual signals D _{i and ω}
C _REG (t) = C (t) / REG (t)
(However,
C (t) = d _{i, ω, 0} × d _{i, ω, t} + ...
+ d _{i, ω, l-1-t} × d _{i, ω, l-1}
And
REG (t) = {(d _{i, ω, 0} ² + ... + d _{i, ω, l-1-t} ² )
× (d _{i, ω, t} ² + ... + d _{i, ω, l-1} ² )} ^0.5
It is. )
If C _REG (t) has a maximum value greater than 0.5, for example, it is determined that it is not noise, and if C _REG (t) does not have a maximum value greater than 0.5, it is noise. It is a method of discrimination.

Various things can be considered for the calculation of the gains Gi _{and ω} performed in step S441, and a preferable example is as follows. That is,
G _{i, ω} = 10 × log ₁₀ {Avg (d _{i, ω} ² )},
Avg (d _{i, ω} ² )
= (d _{i, ω, 0} ² + ... + d _{i, ω, l-1} ² ) / l
It is. The reason why the logarithm is taken is that the relationship between the loudness of the sound and the sensitivity of human hearing is taken into consideration.

In step S447, it is determined whether or not ω has reached a predetermined natural number ω _fin indicating the final band. If not reached (step S447; No), ω is increased by 1 (step S449), and then the process returns to step S437. If reached (step S447; Yes), the process proceeds to step S451. The bands are basically numbered from the low frequency side. That is, ω = 1 means the lowest frequency band, and ω = ω _fin means the highest frequency band.

  In step S451, it is determined whether i has reached M-1. If not reached (step S451; No), i is increased by 1 (step S453), and the process returns to step S433. When it has reached (step S451; Yes), the process proceeds to step S461 in FIG. At this time, a flag and a gain are generated only for the noise band and stored in the storage unit 325. Only the non-noise band components among the components of each band of the residual signal are stored in the storage unit 325.

  In step S461, the CPU 321 sets the input signal sample counter i to i = 0.

The CPU 321 places the non-noise residual signal D _{i, v} = {d _{i, v, 0} ,..., D _{i, v, l−1} } in a general-purpose register, and d _{i, v, 0} ,. , D _{i, v, and l−1} are all initialized to 0 (step S463). Further, the band identification variable ω is set to ω = 1 (step S465).

The CPU 321 searches the inside of the storage unit 325 and investigates whether Di _{and ω} are stored (step S467). As a result of such investigation (step S469), when it is determined that Di _{and ω} are stored (step S469; Yes), the band ω is a non-noise band, and superposition for obtaining a non-noise residual signal. Processing is required. This processing corresponds to the function of the non-noise residual signal totalizer 161 in FIG. Therefore, D _{i, ω} is loaded into the register (step S471) and is superimposed on D _{i, v} stored in the register so far. After Di _{and v} are updated to the numerical sequence obtained as a result of such superposition (step S473), the process proceeds to step S475.

On the other hand, when it is determined in step S469 that Di _{and ω} are not stored (step S469; No), since the band ω is a noise band, the superposition for obtaining the non-noise residual signal is not performed. Then, the process proceeds to step S475.

In step S475, it is determined whether or not ω has reached ω _fin . If it is determined that it has not been reached (step S475; No), ω is incremented by 1 (step S477), and the process returns to step S467 to search for the non-noise band and the above-described case where the non-noise band is found. The superposition process is repeated. When it is determined that ω has reached ω _fin (step S475; Yes), the process proceeds to step S479.

In step S479, the CPU 321 functions as the spectrum converter 163 (FIG. 1) to frequency-convert D _{i and v} by a predetermined method to obtain a spectrum F [D _{i, v} ]. F [D _{i, v} ] is composed of frequency conversion coefficients.

  In the following description and the drawings after FIG. 8, in order to avoid complicated terms, a set of frequency conversion coefficients as a spectrum or a group of frequency conversion coefficients forming a part of the spectrum May be simply referred to as a frequency conversion coefficient. Further, individual frequency coefficients constituting the whole or a part of the spectrum may be referred to as components.

The CPU 321 sets the band identification variable ω to ω = 1 (step S481 in FIG. 9), searches the storage unit 325 to check whether Di _{and ω} are stored (step S483), The presence or absence of storage is determined (step S485). This procedure (steps S481 to S485) is exactly the same as steps S465 to S469, which have already been described, in that it is a procedure related to processing divided into a noise band and a non-noise band.

In this case classification, the CPU 321 determines whether or not the flags Flag _{i and ω} are not found instead of investigating whether or not D _{i and ω} are found as described above when searching inside the storage unit 325. You may investigate. This is because D _{i, ω} and Flag _{i, ω} are alternatively stored in the storage unit, as is clear from the foregoing steps S437 to S449.

If it is determined that D _{i and ω} are stored (step S485; Yes), it means that ω is a non-noise band, so that the CPU 321 functions as the non-noise band extractor 162. , F [D _{i, v} ] to frequency conversion coefficient F [D _{i, v} ] (ω) = {f _{D, i, v, ω 1} ,..., F _{D, i, v, ω , P (ω)} } are cut out (step S487) and stored in the storage unit 325 (step S489). Here, p (ω) is the number of frequency conversion coefficients in the band ω. Thereafter, the process proceeds to step S491.

On the other hand, D _{i, if omega} is judged not to be stored (step S485; No), omega since is that a noise band, is not performed cutout frequency transform coefficients, in step S491 move on.

In step S491, it is determined whether or not ω has reached ω _fin . If it is determined that it has not been reached (step S491; No), ω is increased by 1 (step S493), and then the process returns to step S483. If it is determined that it has been reached (step S491; Yes), the i-th Since this means that the processing for each band in the time interval has been completed, the process proceeds to step S495.

  In step S495, it is determined whether i has reached M-1. If it is determined that it has not been reached (step S495; No), i is incremented by 1 (step S497), then the process returns to step S463, and if it is determined that it has been reached (step S495; Yes), all times Since it means that the processing in the section has been completed, the entire processing for obtaining a flag or the like from the residual signal is completed.

  Next, the procedure of the process performed by the speech encoding apparatus 113 (FIG. 2) according to the second embodiment will be described with reference to the flowchart shown in FIG.

  The CPU 321 sets the input signal sample counter i to i = 0 (step S511 in FIG. 10).

The CPU 321 loads the input signal sample D _i = {d _{i, 0} ,..., D _{i, l-1} } from the storage unit 325 to the register and sets the band identification variable ω to ω = 1. (Step S513), residual signals D _{i, ω} = {d _{i, ω, 0} ,..., D _{i, ω, l-1} } of the band ω are generated (Step S515).

The CPU 321 determines whether Di _{and ω} are noise (step S517). If it is determined that D _{i and ω} are noises (step S517; Yes), generation of flags Flag _{i and ω} in the band _ω , calculation of gains G _{i and ω} , and storage in the storage unit 325 are performed. Are performed (step S519), the process proceeds to step S523.

On the other hand, when it is determined that Di _{and ω} are not noise (step S517; No), the setting of the Band Elimination Filter 169 (FIG. 2) is set not to delete the band ω (step S521). ), And proceeds to step S523.

  Here, the setting of the Band Elimination Filter 169 is cumulatively performed for each loop with respect to the loop processing for ω shown in Step S525 and the like described later. For example, in the noise discrimination processing in step S517, when it is discriminated that the band 2, the band 5 and the band 6 are not noise among all the bands, the setting of the Band Elimination Filter 169 is finally set to the band 2 , The band 5 and the band 6 are set to be deleted except for a total of three bands.

In step S523, it is determined whether or not ω has reached ω _{fin. If} it is determined that ω has not been reached (step S523; No), after increasing ω by 1 (step S525), the process returns to step S515, When it is determined that it has reached (step S523; Yes), the process proceeds to step S527.

  In step S527, it is determined whether or not i has reached M−1. If it is determined that i has not been reached (step S527; No), i is increased by 1 (step S529), and then the process returns to step S513. , When it is determined that it has been reached (step S527; Yes), the process proceeds to step S531.

  At this time, the CPU 321 almost completes functioning as the flag and gain generation unit 133, and also functions as a non-noise band determiner 167 (FIG. 2) to send a deletion band designation command to the Band Elimination Filter 169. Has been completed.

In step S531, the input signal sample counter i is set to i = 0. Then, the CPU 321 loads the residual signal D _i and passes it through the Band Elimination Filter 169 for which the cumulative setting in step S521 has been completed, whereby the non-noise residual signal D _{i, v} = {d _{i, v, 0} , ..., d _{i, v, l-1} } are generated.

  The Band Elimination Filter 169 may be a function or the like included in the CPU operation program stored in the ROM 323 (FIG. 6), or may be provided separately as hardware.

The CPU 321 further obtains a spectrum F [D _{i, v} ] by frequency-converting the generated non-noise residual signals D _{i, v} (step S533).

  In step S535, the band identification variable ω is set to ω = 1.

In step S537 and step S539, it is determined whether the band ω is a noise band or a non-noise band. In FIG. 2, this step corresponds to the non-noise band determiner 167 sending information related to the non-noise band to the non-noise band extractor 162. In the flowcharts shown in FIGS. 8 and 9, the band-based residual signals Di _{and ω} are searched when determining whether the band is a noise band or a non-noise band. As described above, this determination may be performed by searching for the flags Flag _{i and ω} .

On the other hand, in the procedure shown in FIG. 10, as is clear from FIG. 2 showing the second embodiment corresponding to the figure, the process of storing the band-specific residual signals D _{i and ω} in the storage unit 325 (FIG. There is no step S445). This is because the process of superimposing Di _{and ω} is unnecessary in the second embodiment. Therefore, as the determination method in step S537 and step S539 in FIG. 10, only a method by searching for the flags Flag _{i and ω} can be used. This situation also applies to FIGS. 11 and 12, which are flowcharts corresponding to the third embodiment described later.

Eventually, in step S539, it is determined whether or not the flags Flag _{i and ω} are stored in the storage unit 325. If it is determined that Flag _{i, ω} is not stored (step S539; No), the band ω is a non-noise band, and therefore, from the spectrum F [D _{i, v} ] obtained in step S533. The frequency conversion coefficient F [D _{i, v} ] (ω) = {f _{D, i, v, ω 1} ,..., F _{D, i, v, ω, p (ω)} } in the band ω is cut out. After being stored in the storage unit 325 (step S541), the process proceeds to step S543.

On the other hand, when it is determined that Flag _{i and ω} are stored (step S539; Yes), since the band ω is a noise band, the process proceeds to step S543 as it is.

In step S543, it is determined whether or not ω has reached ω _fin . If it is determined that it has not been reached (step S543; No), ω is increased by 1 (step S545), and then the process returns to step S537. If it is determined that it has been reached (step S543; Yes), the i-th Since this means that the processing for each band in the time interval has been completed, the process proceeds to step S547.

  In step S547, it is determined whether i has reached M-1. If it is determined that it has not been reached (step S547; No), after i is increased by 1 (step S549), the process returns to step S533, and if it is determined that it has been reached (step S547; Yes), all times Since it means that the processing in the section has been completed, the entire processing for obtaining a flag or the like from the residual signal is completed.

  Next, a procedure of processing performed by the speech encoding device 115 (FIG. 3) according to the third embodiment will be described with reference to the flowcharts shown in FIGS.

  First, the input signal sample counter i is set to i = 0 (step S540 in FIG. 11).

The input signal samples D _i = {d _{i, 0} ,..., D _{i, l-1} } are loaded (step S542), the spectrum F [D _i ] is obtained (step S544), and the process proceeds to step S547. . As is clear from the position of the spectrum converter 163 in FIG. 3, the third embodiment is characterized in that frequency conversion is performed at an earlier stage than in the previous two embodiments. As described above, since the frequency conversion is performed before various kinds of preprocessing, the frequency conversion coefficient can be obtained with high accuracy without being affected by the error due to the preprocessing.

  In step S546, the band identification variable ω is set to ω = 1.

In step S548, the frequency conversion coefficient F [D _i ] (ω) = {f _{D, i, ω 1} ,..., F _{D, i, ω, p (ω)} } in the band ω is changed to F [D _i ]. At this point, since it is not yet determined whether the band ω is a noise band or a non-noise band, the frequency conversion coefficient F [D _i ] (ω) is generated for all bands ω. The generated F [D _i ] (ω) is stored in the storage unit 325 (step S550).

In step S552, it is determined whether or not ω has reached ω _fin . If it is determined that it has not been reached (step S552; No), after increasing ω by 1 (step S555), the process returns to step S548, and if it is determined that it has been reached (step S552; Yes), the process returns to step S557. move on.

  In step S557, it is determined whether i has reached M-1. If it is determined that it has not been reached (step S557; No), i is increased by 1 (step S559), and then the process returns to step S542. If it is determined that it has been reached (step S557; Yes), the process returns to step S561. move on.

In step S561, the input signal sample counter i is set to i = 0. In step S563, the residual signal D _i is loaded into the register. In step 565, the band identification variable ω is set to ω = 1. In step S567, the band filter unit 127 (FIG. 3) is input from the input signal sample D _i . ) Of the ω-th band filter in) generates the residual signal D _{i, ω} = {d _{i, ω, 0} ,..., D _{i, ω, l-1} } of the band ω.

Here, it is determined whether or not the residual signals D _{i and ω} are noise (step S569). If it is determined that D _{i and ω} are noise (step S569; Yes), flags Flag _{i and ω} are generated and gains G _{i and ω} are calculated (step S571). Flag _{i, ω,} and G _{i , Ω} is stored in the storage unit 325 (step S573), and then the process proceeds to step S575. If it is determined that D _{i and ω} are not noise (step S569; No), the process immediately proceeds to step S575.

In step S575, it is determined whether or not ω has reached ω _fin . If it is determined that it has not been reached (step S575; No), ω is incremented by 1 (step S577), and then the process returns to step S567. If it is determined that it has been reached (step S575; Yes), the process returns to step S579. move on.

  In step S579, it is determined whether i has reached M-1. When it is determined that it has not been reached (step S579; No), i is increased by 1 (step S581), and then the process returns to step S563. When it is determined that it has been reached (step S579; Yes), FIG. The process proceeds to step S591.

  In step S591, the input signal sample counter i is set to i = 0. In step S593, the band identification variable ω is set to ω = 1.

The inside of the storage unit 325 is searched, and it is investigated whether or not the flags Flag _{i and ω} are stored (step S595), and then in the determination step of whether or not Flag _{i and ω} are stored (step S597). move on.

When it is determined that Flag _{i and ω} are not stored (step S597; No), the frequency conversion coefficient F [D _i ] (ω) = {f _{D, i, ω 1} ,. f _{D, i, ω, p (ω)} } are loaded into the register (step S599). In addition to the F [D _i ] (ω), the register includes frequency conversion coefficients F [D _{i, v} ] (ω) = {f _{D, i, v, ω 1} ,. _{f D, i, v, ω} , p (ω)} are _{prepared, F [D i, v]} (ω) = F [D i] by _{(ω) F [D i,} v] (ω) is determined (Step S601).

Here, as described above, F [D _i ] (ω) is generated for all bands, whereas F [D _{i, v} ] (ω) defined in step S601 is determined in step S597. Note that because of the case separation in, it is generated only for the non-noise band. After F [D _{i, v} ] (ω) is stored in the storage unit 325 (step S603), the process proceeds to step S605.

On the other hand, if it is determined that Flag _{i and ω} are stored (step S597; Yes), the process directly proceeds to step S605.

In step S605, it is determined whether or not ω has reached ω _fin . If it is determined that it has not been reached (step S605; No), ω is increased by 1 (step S607), and then the process returns to step S595. If it is determined that it has been reached (step S605; Yes), the process returns to step S609. move on.

  In step S609, it is determined whether i has reached M-1. If it is determined that it has not been reached (step S609; No), i is incremented by 1 (step S611), and the process returns to step S593. If it is determined that it has been reached (step S609; Yes), the process is terminated. To do.

(Procedure for restoring residual signal from flags, etc.)
Hereinafter, a procedure for obtaining a flag, a gain, and a frequency conversion coefficient from a residual signal performed by the speech decoding device 211 according to the fourth embodiment and the speech decoding device 213 according to the fifth embodiment will be described.

As a premise, the decoding unit 223 (FIG. 4 and FIG. 5) performs prediction coefficients, flags Flag _{i, ω} (0 ≦ i ≦ M−1, 1 ≦ ω ≦ ω _fin , where ω is a noise band. ), Gain G _{i, ω} (0 ≦ i ≦ M-1, 1 ≦ ω ≦ ω _fin , where ω is a noise band), frequency conversion coefficient F [D _{i, v} in non-noise band ] (ω) = {f _{D, i, v, ω 1} ,..., f _{D, i, v, ω, p (ω)} } (0 ≦ i ≦ M-1, 1 ≦ ω ≦ ω _fin , Where ω is a non-noise band.) Is already decoded and stored in the storage unit 325.

  First, the procedure of the process performed by the speech decoding apparatus 211 (FIG. 4) according to the fourth embodiment will be described with reference to the flowcharts shown in FIGS.

  First, in step S621 (FIG. 13), the input signal sample counter i is set to i = 0.

The CPU 321 prepares the restored noise residual signal D ′ _{i, uv} and the non-noise residual signal spectrum F [D _{i, v} ] in the register, the component of D ′ _{i, uv} , and F [D _{i , V} ] are all initialized to 0 (step S623).

Note that the initial values of all the components of F [D _{i, v} ] are generally set to 0 as described above. The numerical value 0 is determined from the viewpoint that an offset is unnecessary because the residual signal in the non-noise band is separately restored based on the gain, and is not an absolute rule.

For example, in the first to fifth embodiments, the process related to the gain is omitted, that is, only the flag is transmitted between the speech coding apparatus and the speech decoding apparatus (in other words, between the transmitter and the receiver) as information about the noise band. In this case, the initial value of the component of F [D _{i, v} ] may be set to a predetermined constant other than 0 in consideration of human auditory characteristics. In addition, the predetermined constant may be different for each component. In this way, as will be described later, since the component replacement is performed only for the non-noise band, in the finally generated F [D _{i, v} ], the component in the noise band is equal to the predetermined constant. Remains. In other words, the predetermined constant is a gain determined in advance when no gain is exchanged between the transmitter and the receiver.

After the band identification variable ω is set to ω = 1 (step S625), the inside of the storage unit 325 is searched to check whether or not the flags Flag _{i and ω} are stored (step S627), and the determination step. The process proceeds to (Step S629).

If it is determined that Flag _{i and ω} are stored (step S629; Yes), the band ω is a noise band, and therefore the gain G _{i and ω} should be stored in the storage unit 325. It is. Therefore, G _{i and ω} are loaded into the registers (step S631). In FIG. 4, this is equivalent to the fact that what is passed from the decoding unit 223 to the decoding G switch unit 227 as the band ω information is not a frequency conversion coefficient but a gain.

The CPU 321 functions as the noise sequence generation unit 229 (FIG. 4) for each band _, and generates noise sequences D ′ _{i, uv, and ω} in the band ω using the loaded G _{i and ω} as a clue (FIG. 13). Step S633). A specific generation method will be described later with reference to FIG.

CPU321 is generated D _{i 'i, uv,} the _omega, D is stored in the _register', superimposed on the _uv, it generates a new D _{'i, uv.} That is, D ′ _{i and uv} are updated (step S635 in FIG. 13). In FIG. 4, the noise sequence generated by the band-specific noise sequence generation unit 229 (FIG. 4) is superimposed in the residual signal recovery unit 233, and the residual signal in the noise band is recovered. Equivalent to. When D ′ _{i, uv} is updated as described above, the process proceeds to step S641.

On the other hand, when it is determined in Step S629 that Flag _{i and ω} are not stored (Step S629; No), the band ω is a non-noise band, and therefore the storage unit 325 has a frequency in the band ω. The conversion coefficient F [D _{i, v} ] (ω) should be stored. Therefore, F [D _{i, v} ] (ω) is loaded into the register (step S637). In FIG. 4, this is equivalent to the fact that what is handed over from the decoding unit 223 to the decoding G switch unit 227 as band ω information is not a gain but a frequency conversion coefficient.

The CPU 321 loads F [D _{i, v} ] (ω) obtained by loading the component group in the band ω among the components of the non-noise residual signal spectrum F [D _{i, v} ] stored in the register in step S637. F [D _{i, v} ] is updated by replacing with (step S639).

In the above-described step S623, the initial values of the components of F [D _{i, v} ] are all set to 0, so that a part of F [D _{i, v} ] is obtained by the loop processing for ω in step S643 and the like described later. In each loop, F [D _{i, v} ] (ω) is replaced, and F [D _{i, v} ] is finally suitable as a non-noise residual signal spectrum.

  Here, it should be noted that such substitution occurs only for the components in the non-noise band, and the noise band component remains at the initial value of 0.

In FIG. 4, the processing performed in step S637 and step S639 is performed as follows. In FIG. 4, the frequency conversion coefficient totaling and supplementing unit 255 receives the frequency conversion coefficients in the non-noise band from the decoding G switch unit 227 and totals them. This is equivalent to the process of obtaining frequency conversion coefficients in the entire band while supplementing 0 as the coefficients. When F [D _{i, v} ] is updated as described above, the process proceeds to step S641.

The reason why F [D _{i, v} ] is gradually completed after the search for Flag _{i, ω} in step S627 is the same as that of the speech encoding apparatus according to the first to fifth embodiments. This is because it is assumed that no frequency conversion coefficient is exchanged in the non-noise band with the speech decoding apparatus.

  The fact that such transfer is not performed may cause a problem that the CPU 321 is subjected to a search load, but the amount of information transmitted from the speech encoding device as the transmitter to the speech decoding device as the receiver is small. This is in accordance with the object of the present invention. In addition, for a general CPU, the load of searching for the presence / absence of concise information such as a flag is hardly a problem in practice.

  However, as a modification of the first to fifth embodiments, information about a part of the non-noise band may be transmitted in a mode in which the frequency conversion coefficient of the band is set to 0 instead of the flag. This increases the amount of information transmitted between the transmitter and the receiver, but it is possible to reduce the above-described CPU search burden and to omit part of the replacement processing performed in steps S637 and S639. Contributes to speeding up the process. In addition, even if the amount of information transmitted between the transmitter and the receiver increases, only a small numerical value 0 is transmitted, so if there is a margin in the given information transmission amount, such a modification The example may be more efficient.

In step S641, it is determined whether or not ω has reached ω _fin. If not reached (step S641; No), ω is increased by 1 (step S643), and then the process returns to step S627 and has been reached. (Step S641; Yes), the process proceeds to Step S645.

In step S645, CPU 321 may by functioning as a spectral inverter 257 in FIG. 4, non-noise residual signal spectrum F [D _{i, v]} from the non-noise residual signal D _i, obtaining the _v. Then, the CPU 321 obtains a restored residual signal D ′ _i by superimposing D ′ _{i, uv} and D _{i, v} (step S647), and stores D ′ _i in the storage unit (step S649).

  In step S651, it is determined whether i has reached M-1. If it is determined that it has not been reached (step S651; No), i is increased by 1 (step S653), and then the process returns to step S623. If it is determined that it has been reached (step S651; Yes), the process is terminated. .

Hereinafter, a specific procedure for generating the band-specific noise sequence D ′ _{i, uv, ω} in step S633 described above will be described with reference to the flowchart shown in FIG.

First, a basic noise sequence R _i = {R _{i, 0} ,..., R _{i, l-1} } having a size of ± 1 and a time interval of a random number is generated (step S655).

Here, _Ri is generated assuming that the sampling interval is the same as the sampling interval of the original residual signal. Therefore, in practice, the value of each element R _{i, 0} ,..., R _{i, l−1} is either 0, +1, or −1. Moreover, in these element sequences arranged in time order, +1 or -1 appears at random number intervals, and the values of the other elements are 0.

By passing the obtained basic noise sequence R _i through a band-pass filter that extracts the component of the band ω, the basic noise sequence R _{i, ω} = {R _{i, ω, 0} , ..., R _{i, ω} of the band _{ω , L-1} } is generated (step S657).

By multiplying the generated basic noise sequence R _{i, ω} of the band ω by the gains G _{i, ω} loaded in step S631 in FIG. 13, the noise sequence D ′ _{i, uv, ω} = {d ′ _{i, uv, ω, 0} ,..., d ′ _{i, uv, ω, l−1} } are generated (step S659), and the process ends.

  Next, the procedure of processing performed by the speech decoding apparatus 213 (FIG. 5) according to the fifth embodiment will be described with reference to the flowchart shown in FIG.

  First, in step S661, the input signal sample counter i is set to i = 0.

A restored residual signal spectrum F [D ′ _i ] is prepared in the register of the CPU 321 and all its components are initialized to 0 (step S663).

After the band identification variable ω is set to ω = 1 (step S665), the inside of the storage unit 325 is searched to check whether or not the flags Flag _{i and ω} are stored (step S667), and the determination step. The process proceeds to (Step S669).

When it is determined that Flag _{i and ω} are stored (step S669; Yes), the gains G _{i and ω} are loaded into the register (step S671).

The CPU 321 updates F [D ′ _i ] by replacing all the components included in the band ω with G _{i, ω} × unit component for the components of the restored residual signal spectrum F [D ′ _i ]. (Step S673). In FIG. 5, this corresponds to the frequency conversion coefficient in the noise band being transferred from the constant frequency conversion coefficient generation unit 259 for each band to the frequency conversion coefficient totalizer 267. Thereafter, the process proceeds to step S679.

On the other hand, when it is determined that Flag _{i and ω} are not stored (step S669; No), the frequency conversion coefficient F [D _{i, v} ] (ω) in the band ω is loaded into the register (step S675). For the component [D ′ _i ], the component group included in the band ω is replaced with F [D _{i, v} ] (ω), so that F [D ′ _i ] is updated (step S677). The process proceeds to step S679.

In step S679, it is determined whether or not ω has reached ω _fin. If it has not been reached (step S679; No), ω is increased by 1 (step S681), and then the process returns to step S667. (Step S679; Yes), the process proceeds to Step S683.

In step S683, a residual signal D ′ _i is obtained from the restored residual signal spectrum F [D ′ _i ] by inverse transformation. After D ′ _i is stored in the storage unit (step S685), the process proceeds to step S687.

  In step S687, it is determined whether i has reached M-1. If it is determined that it has not been reached (step S687; No), i is increased by 1 (step S689), and then the process returns to step S663. If it is determined that it has been reached (step S687; Yes), the process is terminated. .

(Procedure for audio signal restoration)
Hereinafter, with reference to FIG. 16, a description will be given of the procedure of audio signal restoration performed inside the audio decoding device 211 and the audio decoding device 213 shown in FIGS. 4 and 5. Here, the procedure when MLSA analysis is used as predictive analysis will be described as an example, but the procedure when other predictive analysis such as linear predictive analysis is used is also the same.

The receiving unit 221 (FIGS. 4 and 5) receives a code that is the result of encoding the prediction coefficient of the original speech and passes it to the decoding unit 223 (FIGS. 4 and 5). The decoding unit 223 decodes the delivered code and generates a prediction coefficient, a flag, and the like. These are stored in the storage unit 325. In the case of MLSA analysis, the prediction coefficient is MLSA filter coefficient M _i = {m _{i, 0} ,..., M _{i, p−1} } (0 ≦ i ≦ M−1).

After the input signal sample counter is set to i = 1 (step S711 in FIG. 16), the prediction coefficient M _i is loaded from the storage unit 325 to the internal register of the CPU 321 (step S713). Next, the synthesis inverse filter CIM _i is calculated from the prediction coefficient M _i (step S715). This corresponds to the specification of the synthesizing filter unit 237 being determined by the synthesizing filter calculating unit 235 to which the prediction coefficient is handed over in FIGS. 4 and 5.

Subsequently, the restored residual signal D′ i is passed through the synthesis filter CIM _i obtained in step S715. As a result, the restored digital audio signal S ′ _i = {s ′ _{i, 0} ,..., S ′ _{i, l−1} } (0 ≦ i ≦ M−1) is generated (step S717). The restored digital audio signal S ′ _i is stored in the storage unit 325 (step S719). Subsequently, it is determined whether or not i has reached M-1 (step S721). If not (step S721; No), i is increased by 1 (step S723), and then step S713. Return to. If i has reached M−1 (step S721; Yes), the process is terminated.

(Example of procedure for obtaining MLSA coefficients from cepstrum)
Figure 17 is a cepstrum _{C i = {c i, 0} , ···, c i, (l / 2) -1} MLSA filter coefficients from _{M i = {m i, 0} , ···, m i, p _-1 } is a flowchart illustrating an example of a specific procedure. By performing the calculations shown in steps S811 to S835, the MLSA filter coefficient is obtained. α is a numerical value for approximation, and α = 0.35 is preferable when the audio signal is sampled at 10 kHz. Further, β = 1−α ² . m _i (0 ≦ i ≦ p−1) is initialized to 0.

An example of the configuration of the MLSA filter using the MLSA filter coefficient obtained in this way is shown in FIG. P _{1 to} P ₄ are approximation coefficients, and for example, P ₁ = 0.4999, P ₂ = 0.1067, P ₃ = 0.0117, and P ₄ = 0.0005656 are preferable.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation and application are possible as already mentioned some modification examples in the text. The above-described hardware configuration, block configuration, and flowchart are examples, and are not limited.

  For example, the description has been made assuming that a mobile phone is used as the speech encoding / decoding device 311 shown in FIG. The present invention can also be applied in the same manner. For example, when the present invention is applied to a personal computer, if a voice input / output device, a communication device, or the like is added to the personal computer, it can have the function of a mobile phone as hardware. Then, if a computer program for causing a computer to execute the above-described processing is distributed by a recording medium or communication, the computer is installed and executed on the computer, thereby causing the computer to execute the speech encoding apparatus or the speech according to the present invention. It is also possible to function as a decoding device.

  That is, the said embodiment is for description and does not restrict | limit the scope of the present invention. Therefore, those skilled in the art can employ embodiments in which each or all of these elements are replaced with equivalent ones, and these embodiments are also included in the scope of the present invention.

It is a functional block diagram of the speech coder according to Embodiment 1 of the present invention. It is a function block diagram of the audio | voice coding apparatus which concerns on Embodiment 2 of this invention. It is a function block diagram of the audio | voice coding apparatus which concerns on Embodiment 3 of this invention. It is a function block diagram of the audio | voice decoding apparatus which concerns on Embodiment 4 of this invention. It is a function block diagram of the speech decoding apparatus which concerns on Embodiment 5 of this invention. It is a figure which shows the physical structure of the audio | voice encoding and decoding apparatus which concerns on Embodiment 6 of this invention. It is a figure which shows the flow of a linear prediction analysis or MLSA analysis. It is a figure which shows the first half of the flow of the process which produces | generates a flag etc. from a residual signal in Embodiment 1 of this invention. It is a figure which shows the second half of the flow of the process which produces | generates a flag etc. from a residual signal in Embodiment 1 of this invention. It is a figure which shows the flow of the process which produces | generates a flag etc. from a residual signal in Embodiment 2 of this invention. It is a figure which shows the first half of the flow of the process which produces | generates a flag etc. from a residual signal in Embodiment 3 of this invention. It is a figure which shows the second half of the flow of the process which produces | generates a flag etc. from a residual signal in Embodiment 3 of this invention. It is a figure which shows the flow of a process which decompress | restores a residual signal from the flag etc. in Embodiment 4 of this invention. It is a figure which shows the flow of the process which produces | generates the noise sequence for every band in Embodiment 4 of this invention. It is a figure which shows the flow of the process which decompress | restores a residual signal from the flag etc. in Embodiment 5 of this invention. It is a figure which shows the flow which restore | restores an audio | voice signal. It is a figure which shows an example of the flow of calculation of an MLSA filter coefficient. It is a figure which shows an example of an MLSA filter.

Explanation of symbols

  111... Speech encoding apparatus according to Embodiment 1, 113... Speech encoding apparatus according to Embodiment 2, 115... Speech encoding apparatus according to Embodiment 3, 121. ..A / D conversion unit, 125 ... predictive analysis unit, 127 ... band filter unit, 129 ... noise discrimination unit, 131 ... encoding A switch unit, 133 ... flag and gain Generating unit, 135... Aggregation conversion unit, 137... Encoding unit, 139... Transmitting unit, 141... Predictive analysis inverse filter calculator, 143. Second band filter, 147... First noise discriminator, 149... Second noise discriminator, 151... 1A switch, 153... 2A switch, 155. And a first gain calculator 157 Second flag generation and second gain calculator, 159... Flag and noise gain totalizer, 161... Non-noise residual signal totalizer, 162. Converter 165... Coding C switch section 167... Non-noise band determiner 169... Band Elimination Filter 171... 1C switch 173. ... Frequency conversion coefficient discriminator, 177... E switch unit for encoding, 179... Non-noise frequency conversion coefficient totalizer, 181... 1E switch, 183. Speech decoding apparatus according to the fourth embodiment, 213... Speech decoding apparatus according to the fifth embodiment, 221... Receiving section, 223... Decoding section, 225. For decryption G switch unit, 229... Noise sequence generation unit for each band, 231... Total reverse conversion unit, 233... Residual signal restoration unit, 235. Filter unit, 239 ... D / A conversion unit, 241 ... Speaker, 243 ... First flag presence / absence discriminator, 245 ... Second flag presence / absence discriminator, 247 ... First G switch, 249 ... 2nd G switch, 251 ... 1st noise train generator, 253 ... 2nd noise train generator, 255 ... Frequency conversion coefficient totaling and supplementer, 257 ... Spectral inverse transformer, 259 ... Constant frequency conversion coefficient generation unit for each band, 261 ... Collective tabulation inverse conversion unit, 263 ... First constant frequency conversion coefficient generator, 265 ... Second constant frequency conversion coefficient generator, 267 ... Frequency conversion coefficient totalizer, 31 ... Speech encoding / decoding device according to the sixth embodiment, 321 ... CPU, 323 ... ROM, 325 ... storage unit, 327 ... radio communication unit, 329 ... voice processing unit, 331: Operation key input processing unit, 333: System bus, 335 ... Antenna, 337 ... Operation key, 339 ... RAM, 341 ... Hard disk

Claims (10)

A prediction analysis unit that decomposes a speech signal into a prediction coefficient and a residual signal by prediction analysis;
A residual signal generator for each band that divides the residual signal into residual signals for each band;
A noise discriminating unit that discriminates whether or not the band is a noise band for each band of the residual signal;
A flag generating unit for generating a flag indicating that the band is a noise band for the band determined to be a noise band by the noise determining unit and obtaining a gain of a residual signal for each band of the band ;
Non-noise band conversion for generating a frequency conversion coefficient in a non-noise band by superimposing the band-specific residual signals of the band determined not to be a noise band by the noise determination unit in the real-time domain and then performing frequency conversion And
An encoding unit that encodes the prediction coefficient obtained by the prediction analysis unit, the flag and gain obtained by the flag generation unit, and the frequency conversion coefficient generated by the non-noise band conversion unit ;
A speech encoding device comprising: A prediction analysis unit that decomposes a speech signal into a prediction coefficient and a residual signal by prediction analysis;
A full-band conversion unit that frequency-converts the residual signal to generate a frequency conversion coefficient;
A residual signal generator for each band that divides the residual signal into residual signals for each band;
A noise discriminating unit that discriminates whether or not the band is a noise band for each band of the residual signal;
A flag generating unit for generating a flag indicating that the band is a noise band for the band determined to be a noise band by the noise determining unit and obtaining a gain of a residual signal for each band of the band ;
From the frequency conversion coefficients obtained by the all-band conversion unit, a totaling unit that totalizes the frequency conversion coefficients of the band determined not to be a noise band by the noise determination unit,
An encoding unit that encodes the prediction coefficient obtained by the prediction analysis unit, the flag and gain obtained by the flag generation unit, and the frequency conversion coefficient tabulated by the tabulation unit ;
A speech encoding device comprising: The noise discrimination unit is
Determining whether or not the band is a noise band based on the shape of the autocorrelation function of the band-specific residual signal for each band;
The speech encoding apparatus according to claim 1 or 2 , characterized in that The prediction analysis unit
An MLSA filter coefficient is obtained as the prediction coefficient by MLSA (Mel Log Spectrum Approximation) analysis, and the residual signal is obtained using an inverse filter defined by the MLSA filter coefficient.
The speech coding apparatus according to any one of claims 1 to 3 , wherein The prediction analysis unit
A linear prediction coefficient is obtained as the prediction coefficient by linear prediction analysis, and the residual signal is obtained using an inverse filter defined by the linear prediction coefficient.
The speech coding apparatus according to any one of claims 1 to 3 , wherein Prediction coefficient generated from speech signal by prediction analysis, flag indicating that specific band of residual signal generated from speech signal by prediction analysis is noise band, gain of residual signal by band in noise band And a receiving unit that receives a code that is a result of encoding a frequency conversion coefficient in a non-noise band ;
A decoding unit that decodes the prediction coefficient, the flag, the gain, and the frequency transform coefficient in the non-noise band from the code;
A noise string generation unit that generates a noise string whose amplitude is adjusted by the gain for each band in a band indicated by the flag to be a noise band;
In the band indicated by the flag as a noise band, all the frequency conversion coefficients in the band are stored as 0, and in the non-noise band, the decoded frequency conversion coefficient is stored to store the entire band. An inverse transform unit that generates a frequency transform coefficient and inversely transforms the generated frequency transform coefficient to obtain a residual signal in a non-noise band ;
A residual signal restoration unit that generates a restored residual signal by superimposing a noise sequence generated by the noise sequence generation unit and a residual signal in a non-noise band obtained by the inverse transformation unit ;
A synthesizing unit that generates a restored speech signal by synthesizing the prediction coefficient decoded by the decoding unit and the restored residual signal generated by the residual signal restoring unit ;
A speech decoding apparatus comprising: A predictive analysis step that decomposes the speech signal into predictive coefficients and residual signals by predictive analysis;
A band-specific residual signal generating step of dividing the residual signal into band-specific residual signals;
A noise determination step for determining whether the band is a noise band for each band of the residual signal;
A flag generating step for generating a flag indicating that the band is a noise band for the band determined to be a noise band in the noise determining step, and obtaining a gain of the band-specific residual signal of the band ;
Non-noise band conversion that generates a frequency conversion coefficient in a non-noise band by superimposing the band-specific residual signals in the band determined not to be a noise band in the noise determination step in the real-time domain and then performing frequency conversion Steps,
An encoding step for encoding the prediction coefficient obtained in the prediction analysis step, the flag and gain obtained in the flag generation step, and the frequency conversion coefficient obtained in the non-noise band conversion step ;
A speech encoding method comprising: Prediction coefficient generated from speech signal by prediction analysis, flag indicating that specific band of residual signal generated from speech signal by prediction analysis is noise band, gain of residual signal by band in noise band and a receiving step of receiving a code that is the result of the frequency transform coefficients, but coded in a non-noise band,
Decoding the prediction coefficient, the flag, the gain, and the frequency transform coefficient in the non-noise band from the code;
A noise sequence generating step for generating a noise sequence whose amplitude is adjusted by the gain for each band in a band indicated by the flag to be a noise band;
In the band indicated by the flag as a noise band, all the frequency conversion coefficients in the band are stored as 0, and in the non-noise band, the decoded frequency conversion coefficient is stored to store the entire band. An inverse transform step for generating a frequency transform coefficient and performing a spectrum inverse transform on the generated frequency transform coefficient to obtain a residual signal in a non-noise band ;
A residual signal restoration step of generating a restored residual signal by superimposing the noise sequence generated in the noise sequence generation step and the residual signal in the non-noise band obtained in the inverse transformation step ;
A synthesis step of generating a restored speech signal by synthesizing the prediction coefficient decoded in the decoding step and the restored residual signal generated in the residual signal restoration step ;
A speech decoding method comprising: On the computer,
A predictive analysis step that decomposes the speech signal into predictive coefficients and residual signals by predictive analysis;
A band-specific residual signal generating step of dividing the residual signal into band-specific residual signals;
A noise determination step for determining whether the band is a noise band for each band of the residual signal;
A flag generating step for generating a flag indicating that the band is a noise band for the band determined to be a noise band in the noise determining step, and obtaining a gain of the band-specific residual signal of the band ;
Non-noise band conversion that generates a frequency conversion coefficient in a non-noise band by superimposing the band-specific residual signals in the band determined not to be a noise band in the noise determination step in the real-time domain and then performing frequency conversion Steps,
An encoding step for encoding the prediction coefficient obtained in the prediction analysis step, the flag and gain obtained in the flag generation step, and the frequency conversion coefficient obtained in the non-noise band conversion step ;
A computer program that executes On the computer,
Prediction coefficient generated from speech signal by prediction analysis, flag indicating that specific band of residual signal generated from speech signal by prediction analysis is noise band, gain of residual signal by band in noise band and a receiving step of receiving a code that is the result of the frequency transform coefficients, but coded in a non-noise band,
Decoding the prediction coefficient, the flag, the gain, and the frequency transform coefficient in the non-noise band from the code;
A noise sequence generating step for generating a noise sequence whose amplitude is adjusted by the gain for each band in a band indicated by the flag to be a noise band;
In the band indicated by the flag as a noise band, all the frequency conversion coefficients in the band are stored as 0, and in the non-noise band, the decoded frequency conversion coefficient is stored to store the entire band. An inverse transform step for generating a frequency transform coefficient and performing a spectrum inverse transform on the generated frequency transform coefficient to obtain a residual signal in a non-noise band ;
A residual signal restoration step of generating a restored residual signal by superimposing the noise sequence generated in the noise sequence generation step and the residual signal in the non-noise band obtained in the inverse transformation step ;
A synthesis step of generating a restored speech signal by synthesizing the prediction coefficient decoded in the decoding step and the restored residual signal generated in the residual signal restoration step ;
A computer program that executes

Images