JPH05158495A - Voice encoding transmitter - Google Patents

Abstract

(57) [Summary] [Object] The present invention relates to a voice coding transmission apparatus for transmitting a voice signal by compressing the amount of information by high-efficiency coding, and more particularly to a spectrum of quantization noise using an auditory weighting filter. The present invention relates to a speech coding and transmission device that reduces noise sensation by shaping noise, and improves the perceptual quality of a reproduced signal by always optimizing the auditory weighting filter used to reduce noise sensation. The purpose is to achieve. [Structure] On the transmitting side, a plurality of encoders 101 _{# 1} having auditory weighting filters 100 _{# 1 to} 100 _#n with different characteristics
To 101 and _#n, a plurality of transmitting-side decoder 10 for decoding an output signal of the encoder corresponding provided at a plurality of encoders corresponding
2 # _1-102 and _#n, among the plurality of encoders 101 # ₁ to 101 _#n is evaluated according to the corresponding quantization noise a predetermined evaluation function generated by the encoder on the basis of a reproduction signal from the transmitting decoder Evaluating means 103 for determining an encoder with less noise
And selection means 104 for transmitting the output signal of the encoder with less noise selected according to the judgment result of the evaluation means 103 to the receiving side.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding transmission apparatus for transmitting a voice signal by compressing the amount of information by high-efficiency coding and, more particularly, shaping a spectrum of quantization noise using an auditory weighting filter. The present invention relates to a voice coding / transmission device that reduces the noise sensation.

With the spread of digital communication lines in recent years,
In order to effectively use the line, a method of compressing the information amount of the voice signal and encoding the voice signal with high efficiency has been attracting attention when transmitting the voice signal. Further, such high-efficiency encoding of the voice signal is useful for reducing the capacity of the voice information storage memory not only in the communication line but also in the voice storage response system.

[0003]

2. Description of the Related Art Conventionally, in a voice encoding / transmission apparatus for performing high-efficiency encoding of a voice signal, a spectrum of quantization noise is shaped by using a perceptual weighting filter for the purpose of reducing a feeling of noise in a reproduced signal. It is known to do so. When using such a perceptual weighting filter,
An auditory weighting filter provided only on the transmitting device,
Some are provided only on the receiving side device, or some are provided on both. Documents disclosing a device using this auditory weighting filter include, for example, the following.

(1) MR Schroeder, et.al, "Optimizing digital speech coder by exploting masking propertis of the human ear" J.Acoust.Soc.Amer., Vol.66, pp.1647-1651, Dec 1979. (2) BSet.al, "Predictive coding of speech at low bit rate" IEEE Trans.Commun., Vol.COM-, pp.600-614, Apr.1982. (3) MRSchroeder, et.al. "Code-excited linear prediction (CELP) 9: High -quality speech at very low bit rate" in Proc.IEEE Int.Conf.Acoust., Spee ch, Signal Processing, Tampa, FL, 1985, vol.1, paper 25.1 .1. (4) CCITT G.721 (5) NSJAYAT et.al, "ADAPTIVE POSTFILTERING OF 16Kb / s-ADPCM SPEECH," Proc.ICASSP pp.829-832, April 1986. (6) MRSchroeder et. al "Code-Excited Linear Prediction (CELP): High Quality Speech at Low Bit Rates," Proc.ICASSP, pp937-940, March 1985. (7) IAGerson et.al, "Vector Sum Excited Lineat Predicton (VSELP) Spe ech coding at 8 kbps, "Proc ISASSP, pp461-464, April 1990. p (8) P. Kroon et.al," A Class of Analysis-by-Synthesis Predictive Coder s for High Qualty Spe ech Coding at Rates Between 4.8 and 16 kbit / s, "IEE E JSAC, vol.6, No.2, February. (9) NSJAYANT et.al," Digital Coding of Waveform ", pp.135, Pretence-ha ll

[0005]

In these conventional devices, only one auditory weighting filter is used, and therefore, the filter characteristic determined by the filter type, order, leak coefficient, etc. is of one type. It is a thing.
For this reason, it cannot be said that optimum noise shaping is always performed for the spectrum of the input audio signal, and the sense of noise in the reproduced signal becomes large, often causing deterioration of audio quality in hearing. ..

The present invention has been made in view of the above problems, and an object of the present invention is to optimize the auditory weighting filter used for reducing the feeling of noise at all times to improve the auditory sense of a reproduced signal. It is to improve quality.

[0007]

1 to 3 are explanatory views of the principle according to the present invention. As shown in FIG. 1, the speech coding and transmitting apparatus according to the present invention has, as one mode, a perceptual weighting filter 100 _{# 1} to _{# 1} having different characteristics on the transmitting side.
A plurality of encoders 101 _# 1 to 101 _#n with 100 _#n, a plurality of transmitting-side decoder 102 for decoding an output signal of the encoder corresponding provided at a plurality of encoders correspond _# 1-102
_#n and the quantization noise generated in the corresponding encoder based on the reproduced signal from the transmission side decoder are evaluated according to a predetermined evaluation function, and there is little noise out of the plurality of encoders 101 _{# 1 to} 101 _#n. The evaluation means 103 for judging the encoder and the selection means 1 for transmitting the output signal of the encoder with less noise selected according to the judgment result of the evaluation means 103 to the receiving side.
04 and 04.

In the speech coding and transmitting apparatus of this aspect,
The transmitter decoder can be configured to perform decoding using an auditory weighted inverse filter that has the inverse characteristic of the auditory weighted filter of the corresponding encoder. Further, in the speech coding and transmitting apparatus of this aspect, the transmitting side decoder may be included as a part of the configuration of the corresponding encoder such as a CELP encoder.

Further, the speech coding and transmitting apparatus according to the present invention is
As shown in FIG. 2, as another form, the auditory weighting filters 200 _{# 1} to 200 having different characteristics are provided on the transmitting side.
A plurality of encoders 201 _{# 1 to} 201 _#n having _#n and the magnitude of perceptually weighted quantization noise generated by the encoder based on the characteristics of the quantizer and the auditory weighting filter is used for each encoder 201. Quantization noise estimating means 202 for estimating each of _{# 1 to} 201 _#n , and a plurality of encoders 20 based on the estimation result of the quantization noise estimating means 202 and the input speech signal.
1 _# 1 ~201 _#n more encoders 201 _# 20 1 to 20 with the feeling of noise was evaluated in accordance with a predetermined evaluation function for the input audio signal
An evaluation unit 203 that determines an encoder with less noise out of 1 _#n , and a selection unit that encodes the input speech signal using the encoder with less noise selected according to the determination result of the evaluation unit 203. And 204.

Further, the speech coding and transmitting apparatus according to the present invention comprises:
The transmission side is equipped with multiple encoders having auditory weighting filters with different characteristics, and the quantization noise of these multiple encoders is evaluated according to a predetermined evaluation function to determine the encoder with less noise, and the noise perception is determined. It is configured to transmit the output signal of the encoder having less number of signals to the receiving side.

Further, as shown in FIG. 3 and as another embodiment, the speech coding and transmitting apparatus according to the present invention has a coder 301 for coding an input speech signal on the transmitting side and an output of the coder 301. A transmission side decoder 302 and a transmission side decoder 302, which decode a signal using a plurality of auditory weighting filters 300 _{# 1 to} 300 _#n, which have different characteristics, respectively.
A plurality of auditory weighting filters 300 _{# 1 to} 300 _#n are evaluated by evaluating the noise sensation of the reproduced signal based on the reproduced signal and the input audio signal decoded by
And a selection information transmitting unit 304 for transmitting the selection information regarding the auditory weighting filter with less noise sense determined by the evaluating unit 303 to the receiving side. On the side, a plurality of perceptual weighting filters 400 _{# 1 to} 400 _#n having the same characteristics as the transmitting side but different in characteristics are provided, and the receiving side device selects from among the plurality of perceptual weighting filters 400 _{# 1 to} 400 _#n. The perceptual weighting filter corresponding to the information is selected, and the received signal is decoded using the perceptual weighting filter.

[0012]

In the speech coding transmission apparatus of the form shown in FIG. 1, a plurality of encoders 101 _{# 1} to 10 _{# 1} to 10 _#n having different auditory weighting filters 100 _{# 1 to} 100 _#n on the transmitting side are provided.
1 _#n are operated in parallel to encode the input voice signal, and the encoded signals of the encoders 101 _{# 1 to} 101 _#n are decoded by the corresponding transmission side decoders 102 _{# 1 to} 102 _#n. Then, the reproduced signals are evaluated by the evaluation means 103 to determine which of the encoders 101 _{# 1 to} 101 _{#n is} the encoder that has less noise with respect to the input voice signal. Selection means 104 based on the determination result
Then, the output signal of the encoder with less noise is selected and transmitted to the receiving side.

Further, in the speech coding transmission apparatus of the form shown in FIG. 2, on the transmitting side, the quantization noise estimating means 202 makes the auditory-weighted quantization noise of each encoder 201 _{# 1 to} 201 _#n. Is estimated based on the characteristics of the quantizer of the encoder and the auditory weighting filter. The evaluation unit 203, based on the estimation result of the quantization noise estimation unit 202 and the input speech signal, each of the encoders 201 _{# 1} to _# 2.
The noise sensation of 01 _#n input speech signal is evaluated according to a predetermined evaluation function. The selection means 204 uses this evaluation means 2
An encoder with less noise is selected according to the evaluation result of 03, and the input speech signal is encoded using the encoder.

As a result, the audio signal can always be encoded by the encoder having the optimum auditory weighting characteristic (that is, the noise feeling is small). Further, in this apparatus, when evaluating the noise sensation of each coder, a local decoder for actually decoding the output signal of each coder is not necessary, and this is replaced by the estimation by the quantization noise estimation means 202. Further, since it is not necessary to operate the remaining encoders other than the selected one, the amount of hardware and the amount of software can be reduced.

As described above, the speech coding and transmitting apparatus of the present invention shown in FIGS. 1 and 2 actually locally decodes the noise sensation of a plurality of encoders provided on the transmitting side and having auditory weighting filters having different characteristics. Evaluate based on the reproduced signal or estimated based on the estimated reproduced signal, determine the encoder with less noise,
The output signal of the encoder with less noise is transmitted to the receiving side.

Further, in the speech coding and transmitting apparatus of the form shown in FIG. 3, at the transmitting side, the input speech signal is encoded by the encoder 301, and the encoded signal is further transmitted to the transmitting side decoder 3.
02, decoding is performed by using auditory weighting filters having different characteristics, and respective reproduction signals are generated. These reproduction signals are evaluated by the evaluation means 303 according to a predetermined evaluation function to determine which of the auditory weighting filters is used for the reproduction signal having less noise, and the selection information transmitting means 304 receives the selection information. Transmit to the device side.

On the receiving side, based on the selection information transmitted from the transmitting device side, the received signal is obtained using an auditory weighting filter corresponding to the selection information (that is, one which is judged by the transmitting side to be less noisy). Is weighted while being decoded. As a result, it becomes possible for the receiving side to reproduce the audio signal by using the auditory weighting filter having a characteristic that the sense of noise is always small.

[0018]

Embodiments of the present invention will be described below with reference to the drawings. In the following description, circuit elements denoted by the same reference numerals throughout the drawings represent circuit elements having the same function. FIG. 4 shows a voice coding transmission device as an embodiment of the present invention. In FIG. 4, a transmitter is provided with three encoders 1 that operate in parallel with each other on input speech signals to perform encoding, and are provided corresponding to these three encoders 1 and output their output signals. Three local decoders 2 that perform local decoding to generate a reproduced signal, an evaluation unit 4 that evaluates the quantized noise spectrum of the reproduced signals from these three local decoders 2, and an evaluation result of the evaluation unit 4 The selection unit 3 selects one output signal from the three encoders 1 based on the above and sends it to the transmission path side.

In the following description, when it is necessary to individually distinguish and explain circuits having different signal systems such as the three encoders 1 or the local decoders 2, each system is # 1, # 2. # 3 ... are distinguished, and the subscripts # 1, # 2, # are added to the reference symbols and parameters of the circuit elements of the system.
3 ... is attached to distinguish. For example encoder 1
_{Describe as # 1} , 1 _{# 2} , 1 _{# 3} .

Each encoder 1 is composed of a perceptual weighting filter 5 for the transfer function W (z) and a quantizer 6, and the output signal from the quantizer 6 is sent to the transmission line via the selector 3. It Here, the auditory weighting filters 5 _{# 1} , 5 _{# 2} and 5 _{# 3} of the encoders 1 _#n , 1 _{# 2} and 1 _{# 3} have different transfer functions W.
It has (z) _{# 1} , W (z) _{# 2} , and W (z) _{# 3} . Each local decoder 2 is composed of an inverse quantizer 7 and an auditory weighted inverse filter 8 having a transfer function 1 / W (z), and an output signal from the auditory weighted inverse filter 8 is sent to the evaluation section 4. .. Wherein each local decoder 2 _#n, 2 # _2, 2 # perceptual weighting inverse filter 8 _{# 1} of _3, 8 # _2, 8 # ₃ perceptual weighting filter 5 _{# 1,}
Different transfer functions 1 / W (z) _{# 1} and 1 / W corresponding to the transfer functions W (z) _{# 1} , W (z) _{# 2} and W (z) _{# 3} of 5 _{# 2} and 5 _{# 3} , respectively.
It has (z) _{# 2} and 1 / W (z) _{# 3} .

The quantization noise spectrum evaluation unit 4 calculates the loudness of noise in the locally decoded signals based on the locally decoded signals from the respective auditory weighted inverse filters 8 _{# 1} , 8 _{# 2} and 8 _{# 3} , Encoders of each encoder 1 _{# 1} to 1 _{# 3} are evaluated, and encoder 1 of the signal system with the least noise feeling
_{It judges #n} and outputs a control signal to the selecting unit 3 so that the selecting unit 3 selects the coded signal from the encoder 1 _{#n. At} the same time, selection information indicating which encoder has been selected is displayed by the side. The information is transmitted to the receiving device side.

The receiver comprises three decoders 10, each decoder 10 having an inverse quantizer 12 and a transfer function 1 / W.
It is composed of the perceptual weighting inverse filter 13 of (z). These decoders 10 _{# 1} , 10 _{# 2} and 10 _{# 3} have the same configuration corresponding to the local decoders 2 _{# 1} , 2 _{# 2} and 2 _{# 3} of the transmitter. In addition, the selectors 9 and 11 are provided, and _one of the decoders 10 _{# 1} , 10 _{# 2} , and 10 _{# 3} is selected according to the side information sent from the transmission device to decode the received signal. Is becoming

The operation of the embodiment apparatus will be described below. In the transmitter, three encoders 1 _{# 1} ...
1 _{# 3} receives audio signals, and the auditory weighting filters 5 _{# 1 to} 5 _{# 3} have different transfer functions W (z) _{# 1} to _{# 3} .
Quantizers 6 _{# 1 to} 6 after being auditory weighted by W (z) _{# 3}
Quantized by _{# 3} . The encoded signals of these encoders 1 _{# 1} to 1 _{# 3} are decoded by the corresponding local decoders 2 _{# 1} to 2 _{# 3} ,
The reproduced signals from them are input to the quantization noise spectrum evaluation unit 4.

The quantizing noise spectrum evaluating unit 4 analyzes and evaluates the reproduced noises from the local decoders 2 _{# 1} to 2 _{# 3} by calculating the loudness of the quantizing noise, and the loudness of the quantizing noise is evaluated. Determines the smallest encoder. Since the output signal from the encoder having the smallest loudness of the quantization noise has the least audible noise in reproduction, the quantization noise spectrum evaluation unit 4 selects the output signal of the encoder. Switching control,
The output signal of the encoder is sent to the receiving device via the transmission path. At the same time, the selection information indicating which auditory weighting filter is used to select the encoder is also sent to the receiving device as side information.

Upon receiving this side information, the receiving device switches the selection units 9 and 11 according to the side information. As a result, the decoder 10 _#n having the perceptual weighting inverse characteristic corresponding to the perceptual weighting characteristic encoder 1 _#n selected on the transmitting side is selected. That is, when a signal encoded with the perceptual weighting characteristic W (z) _{# 1} that minimizes the loudness of noise is transmitted on the transmitting side, the perceptual weighting characteristic 1 / W (z) _{# 1} is transmitted on the receiving side. It is for decoding the received signal.

As a result, on the transmitting side, the audio signal can always be encoded by the encoder having the optimum auditory weighting characteristic, and high-efficiency encoding of the audio signal having good auditory characteristics can be performed.

The calculation of the loudness of the quantization noise in the quantization noise spectrum evaluation unit 4 can be performed by the following method, for example. (1) FFT (Fast Fourier Transform) or DFT (Discrete Fourier Transform) by cutting the input signal and the noise signal by applying a window function (Hamming window etc.) for each appropriate time frame
Convert to the frequency domain using. Here, the frequency domain value of the input frequency is X _j , and the frequency domain value of the error signal is n.
_{Let j} . (2) The power S _{i for} each critical band of the input signal and the power N _i for each critical band of the error signal are calculated. S _i = Σx _j ² (j is the range of the i-th critical band) N _i = Σn _j ² (j is the range of the i-th critical band) (3) S _i , N _i and the following formula. A power B _i , Q that takes into account the effect of masking by convolving the function B _i with
Calculate _i . 10log ₁₀ B _i = 15.81 +7.5 (i + 0.474) -17.5 (1 + 0.474) ² ) ^1/2 E _i = S _i * B _i (* is a convolution) Q _i = N _i * B _i (4) Let θ _{i be} the minimum audible value of the signal for each critical band,
Find the minimum audible level M _i of noise in the critical band. M _i = max (S _i 10 ** {(-15.5 + i) / 10, θ _i } where ** is a symbol representing a power operation. (5) Obtaining the loudness of noise for each critical band L _i = [max (N _i −M _i , 0) / {1+ (S _i / N _i ) ^ui }] ^vi ui, vi is different for each critical band, but here ui = 2.0 approximately. , Vi = 0.25 (6) Obtaining the loudness Ln of noise Ln = ΣL _i

Various modifications are possible in carrying out the present invention. FIG. 5 shows the configuration of the transmitter side of the voice encoding and transmitting apparatus as another embodiment of the present invention. In this embodiment, the present invention is applied to an APC (Adaptive Predictive Coding) type speech coding transmission device. In the speech coding transmission device of this system, which auditory weighting filter is used on the transmission side. It is possible to reproduce at the receiving side without transmitting to the receiving side the selection information indicating whether or not the encoder using is selected.

In FIG. 5, reference numeral 21 is a filter for obtaining a prediction error, which has a transfer function of 1-P (z) and obtains a prediction error component by applying a filter of 1-P (z) to the input voice signal. .. Here, the transfer function P (z) is the transfer function of the prediction filter, and this prediction filter removes the redundancy due to the adjacent correlation of the speech signal, and the coefficient of the filter is the linear prediction analysis method for each appropriate time frame. It is decided by.

The prediction error component obtained by the filter 21 is 3
It is input to the encoders 22 arranged in parallel on the table. Each encoder 22 includes subtractors 26 and 29, a quantizer 27, and an inverse quantizer 2
8. It has a well-known configuration including an auditory weighting filter 30. The quantizer 27 quantizes the prediction error component to generate a signal in which the amount of information is compressed. At this time, a quantization error occurs. The dequantizer 28 dequantizes this signal, and the difference between the output of the dequantizer 28 and the prediction error component obtained by the subtractor 29 becomes the quantization error.

The perceptual weighting filter 30 perceptually weights the quantization error and inputs its output signal to the subtractor 26. Each auditory weighting filter 30 shapes a quantized noise component to reduce noise on the sense of hearing. The quantizer output signal is decoded and compared with the input signal to obtain shaped quantization noise. The auditory weighting filters 30 _{# 1} to 30 _{# 1} of the respective encoders 1 _{# 1} to 1 _{# 3} have different transfer functions W (z) _{# 1 to} W (z) _{# 3} of auditory weighting characteristics.
The tap information of each auditory weighting filter 30 _{# 1} to 30 _{# 3 is} the tap information of the auditory weighting filter 30 _#n selected in the previous frame at the joint of frames to the tap information of each auditory weighting filter 30 _{# 1} to 30 _{# 3.} Copied. The configuration of the perceptual weighting filters 30 _{# 1} to 30 _{# 3} will be described in detail later.

_One of the output signals of the encoders 22 _{# 1 to} 22 _{# 3} is selected by the selecting section 24 and sent to the transmission line side, and at the same time, it corresponds to each of the encoders 22 _{# 1 to} 22 _{# 3.} It is input to the local decoders 23 _{# 1} to 23 _{# 3} provided. Each local decoder 23 includes an inverse quantizer 31, an adder 32, and a predictor 33.
It is composed including. The inverse quantizer 31 has the same configuration as the inverse quantizer 28. The predictor 33 has a transfer function P (z),
The tap information of each prediction filter 33 _{# 1 to} 33 _{# 3} is the prediction filter 33 _#n selected in the previous frame at the joint of frames.
Information is copied to the tap information of each prediction filter 33 _{# 1 to} 33 _{# 3} .

The output signals of the local decoders 23 _{# 1} to 23 _{# 3} are input to the noise spectrum evaluation section 25. The noise spectrum evaluation unit 25 calculates the loudness Ln of noise for the reproduced signal of each local decoder, and determines the encoder in which the auditory weighting filter having the smallest loudness Ln of noise is incorporated. The result of the determination is given to the selecting unit 24, which causes the selecting unit 24 to select the output signal of the encoder with the smallest loudness and send it to the transmission path side. The process of selecting the output signal is performed for each appropriate frame.

Next, the prediction filters 33 _{# 1 to} 33 _{# 3} and the auditory weighting filter 30 _{# 1} to
A configuration example of 30 _{# 3 and} an example of parameters when the input signal is sampled at 8 kHz will be described.

The update cycle of the prediction filter 33 is 2 in consideration of using FFT at the time of spectrum analysis of noise.
Power of (every 128 samples). Prediction filter 3
3 is obtained from linear prediction analysis. The order is about 10. The transfer functions P (z) of the prediction filters 33 _{# 1 to} 33 _{# 3} are the same, and are shown below. P (z) = Σa _i z ⁻ⁱ where i = 1, 10 and a _i is a prediction coefficient used in the prediction filter.

The perceptual weighting filter 30 is an all-zero filter, the order is the same as the prediction filter 33, and the coefficient of the filter is the prediction coefficient a of the prediction filter 33 multiplied by the leak coefficient γ. The characteristics of the filter are changed by changing.

The transfer functions W (z) _{# 1} , W (z) _{# 2} , W of the auditory weighting filters 30 _{# 1} , 30 _{# 2} , 30 _{# 3} in each encoder 22 _{# 1} , 22 _{# 2} , 22 _{# 3} (z) Configure _{# 3} differently as follows. Here, γ is a leak coefficient as described above. W (z) _{# 1} = Σγ ^-i a _i z ^-i where γ = 0.9 and i = 1,10 W (z) _{# 2} = Σγ ^-i a _i z ^-i where γ = 0.8 , I = 1,10 W (z) _{# 3} = Σγ ^-i a _i z ^-i where γ = 0.7, i = 1,10

The quantizer 27 is normalized by the standard deviation of the prediction error signal for each frame. A 128-point Hamming window is used as the analysis window, and the number of critical bands is 17.

With the above arrangement, the encoders 22 _{# 1 to} 22
The auditory weighting characteristics of _{# 3} are different, so that the input speech signal can be encoded using different auditory weighting characteristics. The encoded signals are respectively decoded by the local decoders 23 _{# 1} to 23 _{# 3} ,
The noise loudness of the decoded signals is evaluated by the noise spectrum evaluation unit 25, and the encoder having the auditory weighting characteristic with the smallest noise loudness is determined.
The selection unit 24 selects the output signal from the encoder having the low noise loudness according to the result of this determination and sends it to the transmission path side.

The receiving side has a local decoder 2 at the transmitting side.
A decoder having the same configuration as that of No. 3 is prepared, and the received signal is decoded by this.

The method of realizing different auditory weighting characteristics in the above-described apparatus of FIGS. 4 and 5 is not limited to the above-described one, and for example, the format of the auditory weighting filter is changed or the order of the auditory weighting filter is changed. Let
Alternatively, different hearing weighting characteristics can be realized by combining these. This will be described below.

First, an example will be described in which the characteristics are changed by changing the form of the auditory weighting filter.
The configurations (transfer functions) of the auditory weighting filters 30 _{# 1} , 30 _{# 2} , and 30 _{# 3} are changed as follows. Transfer function of perceptual weighting filter 30 _{# 1} : 1-W (z) _{# 1} = 1-Σγ- ⁱ _ai z- ⁱ where γ = 0.8 and i = 1,10. Transfer function of perceptual weighting filter 30 _{# 2} : 1-W (z) _{# 1} = (1-Σγ ₁ ^-i _ai z ^-i ) / (1-Σγ ₂ ^-i _ai z ^-i ) where γ ₁ = 0.9 and γ ₂ = 0.4, i = 1,1
Set to 0. Transfer function of perceptual weighting filter 30 _{# 3} : 1-W (z) _{# 3} = 1 / (1-Σγ- ⁱ _ai z ^-i ) where γ = 0.2 and i = 1,10.

In this case, since the tap information of the perceptual weighting filters is copied at the seams of the frames, the input of the past 10 samples and the output of the past 10 samples are held in all the auditory weighting filters 30.

Next, an example will be described in which the characteristics are changed by changing the order of the auditory weighting filter. Perceptual weighting filter 30 # _1, 30 # _2, 30 # ₃ of the transfer function _{W (z) # 1, W} (z) # so that _2, W (z) _{# 3} to below respectively. W (z) _{# 1} = Σγ- ⁱ _ai z- ⁱ However, γ = 0.8 and i = 1,10. W (z) _{# 2} = Σγ- ⁱ _ai z- ⁱ However, γ = 0.8 and i = 1,8. W (z) _{# 3} = Σγ- ⁱ _ai z- ⁱ However, γ = 0.8 and i = 1,6.

In this case, since the tap information of the perceptual weighting filter is copied at the seam of the frames, the input of the past 10 samples (the data of the maximum number of taps) is held in all perceptual weighting filters.

When combining the above-mentioned perceptual weighting filter formats and order changes, the maximum tap number of input data and output data are held in all perceptual weighting filters.

FIG. 6 shows a voice encoding / transmitting apparatus as another embodiment of the present invention. This example illustrates the present invention, for example CELP as described in reference 3) above.
(Code Excited Linear Predective Coding) This is a case where the present invention is applied to a coded transmission device of a system.

In FIG. 6, the apparatus according to this embodiment has three encoders 40 operating in parallel with respect to an input speech signal, and a quantization error made by each encoder 40 is inputted to the quantum encoder for calculating and evaluating its loudness. Noise spectrum evaluation unit 48,
According to the evaluation result of the quantization noise spectrum evaluation unit 48, the selection unit 49 is configured to select the encoding information (codebook index, gain, etc.) of the encoder having the smallest quantization noise and send it to the receiving side.

Each encoder 40 outputs a codebook 41 for generating encoded information to be a transmission signal, a gain coefficient multiplier 42 for changing the gain of the encoded information from the codebook, a pitch prediction filter 43, and a reproduced signal. Proximity prediction filter 44, subtractor 45 that generates a quantization error by taking the difference between a reproduction signal and an input signal, and quantization error 0 from subtractor 45
The perceptual weighting filter 46 that performs perceptual weighting, the squared error that controls the search of the coded information of the codebook 41 so that the perceptual weighting filter 46 calculates the power of the quantization error after weighting, and the error power is minimized. It is configured to include a minimization circuit 47 and the like.

In the CELP type encoder 40, the subtracter 45 subtracts the difference between the input signal and the result obtained by inputting all combinations of the codebook index and the gain to the pitch prediction filter 43 and the proximity prediction filter 44. Then, the auditory weighting filter 46 applies the auditory weighting characteristic to the difference to select the combination of the codebook index and the gain having the smallest power, and sends it to the receiving side.

Here, each pitch prediction filter 43 _# 1-4
The transfer functions 1 / B (z) of 3 _{# 3} are the same, and B (z) = 1-αz ^L , where L is the pitch period.

The transfer functions 1 / A (z) of the proximity prediction filters 44 _{# 1 to} 44 _{# 3} are also the same, and A (z) = 1-Σa _i z ^-i , where i = 1,10. ..

The gains of the gain coefficient multipliers 42 _{# 1 to} 42 _{# 3} are G _{# 1} , G _{# 2} and G _{# 3} , respectively. The codebook indexes of the codebooks 41 _{# 1 to} 41 _{# 3} are n _{# 1} , n _{# 2} and n _{# 3} , respectively.

Auditory weighting filters 46 _{# 1} , 46 _{# 2} , 4
6 _{# 3} is the following transfer function W (z) _{# 1} , W (z) _{# 2} , W
(z) Use _{# 3} . W (z) _{# 1} = (1-Σγ ₁ ^-i _ai z ^-i ) / (1-Σγ ₂ ^-i _ai z ^-i ), where γ ₁ = 1.0 and γ ₂ = 0.4 Yes, i = 1,10. 0. (z) _{# 2} = (1-Σγ ₁ ^-i _ai z ^-i ) / (1-Σγ ₂ ^-i _ai z ^-i ) where γ ₁ = 1.0 and γ ₂ = 0.6 , I = 1,10. W (z) _{# 3} = (1-Σγ ₁ ^-i _ai z ^-i ) / (1-Σγ ₂ ^-i _ai z ^-i ), where γ ₁ = 1.0 and γ ₂ = 0.8 Yes, i = 1,10.

As a result, in each of the encoders 40 _{# 1} to 40 _{# 3} ,
Audio signals are encoded using different auditory weighting characteristics, and the quantization noise spectrum evaluation unit 48 determines which of the encoders has the smallest quantization noise loudness. Based on the determination result, the selection unit 49 selects the coding information of the encoder with the smallest loudness (that is, the codebook index n and the gain G) and sends it to the receiving side.

FIG. 7 shows a voice encoding / transmitting apparatus as another embodiment of the present invention. The speech coding transmission apparatus of each of the above-described embodiments operates a plurality of encoders having different auditory weighting characteristics at the same time. However, in such a speech coding transmission apparatus, a plurality of encoders currently operating at the same time are always required. It is also necessary, and in order to evaluate the noise characteristics of the encoder by the noise spectrum evaluation unit, a plurality of local decoders for locally decoding the output signal of each encoder are also required. Grows very large.
When these encoders and local decoders are realized by software processing using a microcomputer or DSP,
There is a problem that the processing amount becomes very large. The embodiment of FIG. 7 solves this problem.

In FIG. 7, reference numerals 1 _{# 1} to 1 _{# 3} are encoders having different auditory weighting characteristics, 3 and 427 are selectors, and one of the _three encoders 1 _{# 1} to 1 _{# 3} is used. The audio signal is selected and the audio signal is input to the selected encoder and the encoding result is output. Reference numeral 417 is a quantization error estimator, which is based on the characteristics of the perceptual weighting inverse filter corresponding to the perceptual weighting filter of the encoder and the characteristics of the quantizer (quantization step size, etc.). The estimated quantization error (or quantization noise) is estimated for each of the _three encoders 1 _{# 1} to 1 _{# 3} . 437 is a quantization noise spectrum evaluation unit,
Result estimated by the quantization noise estimating unit 417 (i.e. a quantization error of the encoder 1 _{# 1} to 1 # ₃₎ based on the the input speech signal, the encoder for the audio signal 1 _{# 1} to 1 _{# This is a circuit} that calculates and evaluates the loudness of the quantization noise of ₃ and determines the one with the smallest loudness of the quantization noise (that is, the one with the least noise perception).

The operation of the apparatus of this embodiment will be described below.
For each of the encoders 1 _{# 1} to 1 _{# 3} , the quantization error estimator 417 determines the quantization characteristics of that encoder (such as the quantization step size of the quantizer 6) and the characteristics of the auditory weighted inverse filter (transfer function 1). / W (z) etc.) and the characteristic (quantization error) of the noise signal of the encoder is estimated. This quantization error can be estimated almost uniquely by calculation if the quantization characteristic and the inverse filter characteristic 1 / W (z) are known in advance. The method of this estimation will be described later in detail.

The quantization noise spectrum evaluation unit 437 is
Each encoder 1 _{# 1} to 1 estimated by the quantization error estimation unit 417
_# Receives quantization noise power of _3, based on the input speech signal and those of the quantization noise information, the encoder 1 _{# 1} to 1 # ₃ to each encoder when the audio signal is input 1 _# Calculate and analyze the loudness of quantization noise from ₁ to 1 _{# 3} . Here, the one with the smallest loudness of the quantization noise can be regarded as the one with the least audible noise.

The selection units 3 and 427 receive the switching signal which is the evaluation result from the quantization noise spectrum evaluation unit 437, so that the loudness of the quantization noise is selected from the _three encoders 1 _{# 1} to 1 _{# 3.} A coder using a perceptual weighting filter that minimizes is selected, a voice signal is input to the coder, coding is performed, and the coding result is sent to the transmission path side. At this time, information indicating which auditory weighting filter is used to select the encoder is also multiplexed and transmitted as side information.

On the decoder side, the optimum auditory weighting inverse filter is selected from a plurality of auditory weighting inverse filters and used based on the information indicating which auditory weighting filter has been used, and the received signal is decoded.

As described above, the point that the speech coding and transmitting apparatus of this embodiment is different from the coding and transmitting apparatus of each of the above-mentioned embodiments is that
The quantization error power used in the noise loudness evaluation was actually obtained by the local decoders 2 _{# 1} to 2 _{# 3} in each of the above-described embodiments, whereas in the present embodiment, the quantization error estimation unit 417 is used. It is estimated and used, which eliminates the need for a local decoder on the transmission side and the hardware or software for implementing it. Further, in the present embodiment, a plurality of encoders are not simultaneously operated with respect to the input speech signal, and since the speech signal is input only to the encoder evaluated to have the least noise feeling, the encoding is performed. In particular, when the encoding is realized by software, it is not necessary to perform software processing on the remaining encoders, and the software processing can be reduced.

Method of Estimating Quantization Error The method of estimating the noise signal described above will be described below. This estimation is performed for each inverse filter. (1) Auditory weighted inverse filter (1 / W (z) _{# 1} , 1 / W
(z) _{# 2} , 1 / W (z) _{# 3} ) frequency characteristics are obtained. (2) An interval (sampling frequency / sampling frequency / corresponding to the time frame used when calculating the loudness of noise based on frequency characteristics)
Sampling is performed on the frequency axis at (frame length). (3) The sampled value is convolved with the window used for clipping multiplied by FFT (DFT). (4) The power of the quantization error is calculated from the characteristics of the quantizer. This calculation method is made to be different each according to the characteristics of the quantizer, for example, a quantization error power E ² in the case of linear quantizer and the quantization width ^{Δ, E 2 = Δ 2/} 1
2, and the power per sample is E ² / frame length. (5) Assuming that the frequency of the quantization characteristic is white (flat), the characteristic obtained in (3) is multiplied to obtain the power of the noise signal in the frequency domain.

The method of calculating the loudness of noise is the same as the method of calculating the loudness of quantization noise, but the value n _{j in} the frequency domain of the quantization error signal in step (1) is found by the above estimation method. Value.

FIG. 8 shows another embodiment of the present invention. The audio encoding / transmitting apparatus of this embodiment is intended to reduce the hardware / software of the transmitting side in the same manner as the above-mentioned embodiments with respect to the APC system described in FIG. In this embodiment as well, as in the embodiment of FIG. 5 described above, the information indicating which auditory weighting filter has been selected on the transmitting side need not be transmitted to the receiving side.

In FIG. 8, the filter 21 for obtaining the prediction error, the encoders 22 _{# 1 to} 22 _{# 3} , the selection unit 24, etc. are the same as those described in FIG. The difference is that the apparatus of this embodiment is different from the quantization error estimation unit 51 and the encoders 22 _{# 1 to} 22 _{# 3.}
The noise spectrum evaluation unit 52 has a selection unit 53 for switching the input of a voice signal to the noise spectrum evaluation unit 52, and does not have a local decoder.
Is input with the quantization error power estimated by the quantization noise estimation unit 51.

The encoders 22 _{# 1 to} 22 _{# 3 use} three types of auditory weighting filters 30 _{# 1} to 30 _{# 3} . The characteristics of these auditory weighting filters 30 _{# 1} to 30 _{# 3} are W _{# 1} respectively.
(z), W _{# 2} (z) and W _{# 3} (z). These are for shaping the quantization noise component and reducing the audible noise.
The shaped noise components are added to the quantization noise (1-W
(z) _{# 1} ) / (1-P (z)), (1-W (z) _{# 2} ) / (1-
P (z)), and (1-W (z) _{# 3} ) / (1-P (z)) characteristics filters 60 _{# 1 to} 60 _{# 3} are applied. .

In the quantization error estimator 51, for each of the encoders 22 _{# 1 to} 22 _{# 3} , the filter 6 corresponding to them is
From the frequency characteristics of 0 _{# 1 to} 60 _{# 3 and} the characteristics of the quantizer, the power of the perceptually weighted quantization error signal is obtained.
The quantization error information of each encoder is passed to the noise spectrum evaluation unit 52.

The noise spectrum evaluation unit 52 calculates and evaluates the loudness of noise from the quantization error power obtained by the quantization error estimation unit 51 and the input speech signal, evaluates them, and evaluates the loudness of the most noise. The selection unit 53, 24 is configured to determine an encoder combined with a small auditory weighting filter and select the encoder based on the evaluation result.
Control signal to.

The selectors 24 and 53 switch to select the encoder selected by the noise spectrum evaluation unit 52,
As a result, the voice signal is input to the encoder for encoding and the encoded result is output to the transmission path. The process of selecting this output is performed for each appropriate frame. As the tap coefficient of each auditory weighting filter, the tap coefficient of the auditory weighting filter selected at the seam of the frame is copied to the tap coefficient of each auditory weighting filter.

Prediction filter 33 _{# 1} ...
The configuration of 33 _{# 3} and the perceptual weighting filters 30 _{# 1} to 30 _{# 3} can be the same as that described in FIG.

Quantization Error Estimation Method A noise signal estimation method in this embodiment will be described below. (1) (1-W (z) _{# 1} ) / (1-P (z)), (1-W
(z) _{# 2} ) / (1-P (z)), (1-W (z) _{# 3} ) / (1-
The frequency characteristics of the P (z) filter 60 _{# 1 to} 60 _{# 3} are obtained. (2) Sampling is performed every 8000/128 Hz on the frequency axis. (3) Concerning the sampled value, a window used for clipping is convolved with a 128-point Hamming window multiplied by FFT to obtain the characteristics of the filter on the discrete frequency axis. (4) The size of the quantization error is calculated from the characteristics of the quantizer. For example, in the case of a 2-bit non-linear quantizer optimized for the Laplace distribution, the power of the prediction error signal ((1
-P (z)) filtered) to 7.54d
A value smaller than B is the quantization error power. (See reference (3)). (5) Assuming that the frequency characteristic of the quantization error is white (flat), the characteristic obtained in (3) is multiplied to obtain the power of the noise frequency in the frequency domain.

FIG. 9 shows another embodiment of the present invention. In this embodiment, the present invention is applied to a speech coding and transmitting apparatus of a conventional form in which the auditory weighting filter is used only on the decoder side on the receiving side to shape the spectrum of the quantization noise and reduce the feeling of noise. In particular, the present invention relates to ADPCM (Adap as described in the above-mentioned reference (4)).
A post-filter based on a formant structure as described in the above-mentioned document (5) is used as a perceptual weighting filter, which is applied to a high-efficiency voice coding transmission device of a tive differential pulse code modulation (Post-Filter). An example of the case of use will be described.

In FIG. 9, the device on the transmission side is ADPC.
An M encoder 70 and an ADPCM decoder 71 that decodes the encoded information of the ADPCM encoder 70 to generate a reproduction signal.
And three post-processing filters 72 _{# 1 to} 72 _{# 3} based on the formant structure for weighting the reproduced signal from the ADPCM encoder 71 with different characteristics, and these post-processing filters 72 _{# 1 to} 72 _{# 3.} And a quantization noise spectrum evaluation unit 73 that evaluates the quantization noise spectrum of the reproduction signal weighted by the auditory sense.

The device on the receiving side has the same characteristics as the ADPCM decoder 74 which decodes the received signal received from the transmission line to generate the reproduced signal and the post-processing filters 72 _{# 1 to} 72 _{# 3} on the transmitting side. Three post-processing filters 7 for weighting
6 _{# 1 to} 76 _{# 3} and these post-processing filters 76 _{# 1 to} 76 _{# 3}
Of the ADPCM decoder 74, and selects the auditory weight of the reproduced signal from the ADPCM decoder 74.

Here, the post-processing filters 72 _{# 1 to} 7 of the transmitting side
2 _{# 3} and post-processing filters on receiving side 76 _{# 1 to} 76 _{# 3} are ADP
This is realized by multiplying the prediction filter of the CM decoder by the leak coefficients α and β. For example, in the ACPCM decoder described in the above-mentioned document (5), the transfer function of the prediction filter is {1 + B (z)}. / {1 −A (z)} A (z) = Σa _j z ^-j where Σ is j = 1 to 2 B (z) = Σb _j z ^−j However, Σ is j = 1 to 6 And the transfer function of the post-processing filter is {1 + B '(z)} / {1-A' (z)} A '(z) = Σa _j α ^j z ^-j where Σ is j = 1 to 2 O ≦ α ≦ 1 B ′ (z) = Σb _j β ^j z ^−j However, Σ is realized in the form of O ≦ β ≦ 1 from j = 1 to 6.

Here, it is possible to change the characteristics of the perceptual weighting filter by the values of the leak coefficients α and β.
In this embodiment, three kinds of combinations of the values of α and β are set. This combination may be, for example, α = 0.2 β = 0.8 α = 0.1 β = 0.8 α = 0.1 β = 0.9.

The operation of the apparatus of this embodiment will be described. In the transmitting side apparatus, the signal encoded by the ADPCM encoder 70 is decoded by the ADPCM decoder 71, and the reproduced signals thereof are respectively post-processed filters 72 _{# 1 to} 72 _#. Weight the hearing through ₃ . The perceptually weighted reproduction signals are input to the quantized noise spectrum evaluation unit 73, where the loudness of noise is calculated and analyzed between the reproduction signal and the input audio signal, which are shaped with these different perceptual weighting characteristics. Then, from among them, the one with the smallest loudness of the (quantization) noise is determined to have the best auditory characteristic (that is, the least noise feeling), and which auditory weighting filter (that is, the post-processing filter 72 _# The selection information of _{1 to} 72 _{# 3} ) is multiplexed with the coding information of the ADPCM encoder 70 and sent to the receiving side.

The method of calculating the loudness of the noise in the quantization noise spectrum evaluation unit 73 is the same as in the case of the embodiment of FIG. 4 described above, and the three post-processing filters 72 _{# 1 to} 72 _# 72.
_The operation of selecting the one with the best auditory characteristic of _{# 3} is performed for each appropriate frame, and the tap coefficient and the filter coefficient of the selected post-processing filter 72 _#n are selected at the joints of the frames. Copied to the tap and filter coefficients of 72 _{# 1 to} 72 _{# 3} .

In the receiving side device, the selection information is separated from the received signal, and when the post-processing filter of which auditory weighting characteristic is used for the encoded signal which is currently received, the sense of noise of the reproduced signal is the most. It recognizes whether the amount of noise decreases, and switches the selectors 75 and 77 so as to select the post-processing filter 76 _#n that has the least noise sensation, and performs auditory weighting on the decoded signal.

FIG. 10 shows another embodiment of the present invention. In this embodiment, like the embodiment of FIG. 9 described above, the present invention is applied to a voice coding transmission apparatus in which noise is shaped by performing auditory weighting only on the receiving side apparatus. Similarly, the ADPCM method is used as the encoding method and the post-processing filter based on the formant structure is used as the perceptual weighting filter.

The difference from the embodiment of FIG. 9 is that the quantization noise spectrum evaluation unit 7 calculates the noise loudness.
That is, the perceptual weighting filter is applied to the input audio signal input to No. 3. That is, the post-processing filter 72
Three post-processing filters 78 _{# 1 to} 78 _{# 3} having characteristics corresponding to _{# 1 to} 72 _{# 3} and based on a formant structure as a perceptual weighting filter are provided, and the perceptual weighting is applied to the input speech signal to quantize them. It is input to the digitized noise spectrum evaluation unit 73.

In the quantization noise spectrum evaluation section 73,
The noise loudness is calculated between the signal obtained by applying the post-processing filter 78 _{# 1} to the input audio signal and the signal obtained by applying the post-processing filter 72 _{# 1} to the reproduction signal decoded by the ADPCM decoder 71. Similarly, between the signal obtained by applying the post-processing filter 78 _{# 2} to the input audio signal and the signal obtained by applying the post-processing filter 72 _{# 2} to the reproduction signal decoded by the ADPCM decoder 71 The noise loudness is calculated between the signal applied with the processing filter 78 _{# 3} and the signal obtained by applying the post-processing filter 72 _{# 3} to the reproduced signal decoded by the ADPCM decoder 71, respectively. Then, of these, the post-processing filter having the smallest noise loudness is determined, and the selection information is transmitted to the receiving side device.

The receiving side device has the same configuration as that described with reference to FIG. 9, and its operation is also the same.

In the embodiment of FIG. 10, the post-processing filter 7
2 _{# 1 to} 72 _{# 3} has a feature that a relatively appropriate auditory weighting characteristic can be selected when the auditory weighting is strong and the reproduced signal feels like a muffled sound.

11 and 12 show another embodiment of the present invention. The speech coding transmission apparatus of this embodiment is obtained by applying the present invention to a speech coding transmission apparatus of the type that performs noise shaping by performing auditory weighting only on the receiving side apparatus, as in FIGS. 9 and 10 described above. The CELP method is used as the encoding method, and the pitch pre-processing filter (Pitch-prefilter) based on the pitch structure as described in the above-mentioned document (7) is used as the auditory weighting filter.

FIG. 11 shows an embodiment of the transmitting apparatus, in which 80 is a CELP type encoder. 81-
Reference numeral 85 is a circuit for decoding the information encoded by the CELP encoder 80 on the transmission device side, 81 is a codebook, 82 is a gain coefficient multiplier, and 83 _{# 1 to} 83 _{# 3} have different hearing characteristics. Three pitch pre-processing filters based on pitch structure as weighting filters, 84 _{# 1 to} 84
_{# 3} is a pitch prediction filter to which the output signals of the pitch pre-processing filters 83 _{# 1 to} 83 _{# 3} are input, and 85 _{# 1 to} 85
_{# 3} is a proximity prediction filter to which the output signals of the pitch prediction filters 84 _{# 1 to} 84 _{# 3} are input, and 86 is an output signal of the proximity prediction filters 85 _{# 1 to} 85 _{# 3} (that is, a decoded reproduction signal). Are input, and the quantization noise spectrum evaluation unit evaluates the quantization noise spectrum of each reproduced signal based on them and the input audio signal.

FIG. 12 shows an embodiment of the receiving apparatus. 91 is a codebook, 92 is a gain coefficient multiplier, and 93 _{# 1 to} 93 _{# 3} are pitch structures as auditory weighting filters having different characteristics. 3 pitch pre-processing filters based on this pitch pre-processing filter 93 _{# 1}
~ 93 _{# 3} is the pitch pre-processing filter 83 _{# 1} on the transmitter side
It has the same characteristics corresponding to 83 _{# 3} .
Reference numeral 96 denotes a selection unit that selects one of the pitch preprocessing filters 93 _{# 1 to} 93 _{# 3} , 94 denotes a pitch prediction filter to which the output signal of the pitch preprocessing filter 93 _#n selected by the selection unit 96 is input, 95 Is a proximity prediction filter that receives the output signal of the pitch prediction filter 94 and generates a decoded reproduction signal.

In this embodiment, the pitch prediction filter 84,
The transfer function B (z) of 94 is set to B (z) = 1-αz- ^L (where L is a pitch period and α is a constant such as 0.8). Also, the proximity prediction filter 85,
The transfer function A (z) of 95 is set to A (z) = 1-ΣA _j z ^-j (where Σ is from j = 1 to N, and N is the prediction order). Further, G is a gain coefficient multiplier 82, 9
A gain of 2 and n are codebook indexes of the codebooks 81 and 91.

In this embodiment, the pitch preprocessing filter 8
Three different characteristics are realized by making the characteristics of 3 _{# 1 to} 83 _{# 3} as follows. Transfer function of pitch pre-processing filter 83 _{# 1} = 1 / (1-β ₁ z ^-L ) Transfer function of pitch pre-processing filter 83 _{# 2} = 1 / (1-β ₂ z ^-L ) Pitch pre-processing filter 83 _# Transfer function of ₃ = 1 / (1-β ₃ z ^-L ) where β ₁ , β ₂ , and β ₃ are values β obtained from the input speech signal
Is multiplied by a different coefficient. Of course, in addition to this, different auditory weighting characteristics are realized by fixedly providing β ₁ , β ₂ , and β ₃ or changing the value of the pitch period L by the pitch preprocessing filter. be able to.

The operation of the apparatus of this embodiment will be described. On the transmitting apparatus side, information encoded by the CELP encoder 80 is processed by pitch preprocessing filters (auditory weighting filters) 83 _{# 1 to} 83 _{# 3} having different characteristics. Each of them is decoded and the reproduced signals thereof are input to the quantization noise spectrum evaluation unit 86, and the noise noise is minimized from the decoded reproduction signals in the quantization noise spectrum evaluation unit 86. The pitch pre-processing filter 83 _#n corresponding to is determined. Then, the selection information indicating which pitch pre-processing filter it is is transmitted to the receiving device side.

On the receiving device side, it is known which pitch pre-processing filter is used to decode the received signal to decode the received signal to reduce the noise, and the pitch pre-processing filter 93 _{#n is used.} Is selected by the selection unit 96 to decode the received signal.

FIG. 13 shows the structure of the voice coding transmission device on the side of the transmitter as still another embodiment of the present invention. The voice coding transmission device of this embodiment is the same as that shown in FIG.
A speech coding transmission apparatus of a type in which noise is shaped by performing auditory weighting only in the receiving side apparatus as in 2, and the CELP method is used as an encoding method, and the auditory weighting filter is described, for example, in the above-mentioned document (8). Post-processing filter (Pitch-
postfilter) is used.

In FIG. 13, reference numeral 80 is a CELP type encoder, 81 is a codebook, 82 is a gain coefficient multiplier,
84 is a pitch prediction filter, 85 is a proximity prediction filter,
Reference numeral 86 denotes a quantization noise spectrum evaluation unit, which are the same as those described in FIG. 87 _{# 1 to} 87 _{# 3} are pitch post-processing filters based on the pitch structure, function as perceptual weighting filters, and have different perceptual weighting characteristics. For example, pitch post-processing filter 87 _{# 1} = 1 (1-ε ₁ z ^-L ) pitch post-processing filter 87 _{# 2} = 1 (1-ε ₂ z ^-L ) pitch post-processing filter 87 _{# 3} = 1 (1- ε ₃ z ^-L ). Here, the other parameters are the same as in the case of the embodiment shown in FIG.

Although not shown, the configuration on the receiving device side is similar to that of the decoding unit on the transmitting side.
Gain coefficient multiplier 82, pitch prediction filter 84, proximity prediction filter 85, pitch post-processing filters 87 _{# 1 to} 87
It is configured to include _{# 3} and further includes a selection unit that selects one of the pitch post-processing filters 87 _{# 1 to} 87 _{# 3} according to the selection information from the transmission device side and outputs it as a reproduction signal on the reception side. is there.

The operation of the apparatus of this embodiment is similar to that of the embodiment of FIG. 11, and the information encoded by the CELP encoder 80 on the transmitter side is processed by pitch post-processing filters (auditory weighting) having different characteristics. Filter) 87 _{# 1} ~
87 _{# 3} , each of them is decoded, and the reproduced signals thereof are input to the quantization noise spectrum evaluation unit 86. In this quantization noise spectrum evaluation unit 86, the noise loudness from the decoded reproduction signal is increased. The pitch post-processing filter 87 _#n corresponding to the smallest one is determined, and selection information indicating which pitch post-processing filter it was is transmitted to the receiving device side.

The receiving apparatus side knows which pitch post-processing filter should be used to decode the received signal to decode the received signal to reduce the noise, and the pitch post-processing filter 97 _#n To decode the received signal.

The embodiments of FIGS. 11 and 13 can be modified in various ways. For example, in the embodiment of FIG. 11, it is possible to provide a post-processing filter based on the formant structure after the proximity prediction filters 85 _{# 1 to} 85 _{# 3} . These post-processing filters may have the same characteristics or may have different characteristics. This enables auditory weighting by both the pitch structure and the formant structure.

Also in the embodiment shown in FIG. 13, one post-processing filter based on the formant structure is provided after the proximity prediction filter 85, and its output signal is branched into three to obtain pitch post-processing filters 87 _{# 1 to} 87 _{#, respectively.} You may input in ₃ . In addition, various combinations of filters are possible.

In the above embodiment, the auditory weighting filter has three characteristics in order to simplify the description, but the present invention is not limited to this, and may be three or more, or two. May be.

The method of implementing the plurality of encoders on the transmission side in the above-described embodiment may be a hardware circuit or a software process. Further, one encoder is prepared as a hardware circuit, and various encoders are equivalently realized by setting various parameters such as the auditory weighting characteristic function and the quantization step size by software. May be

[0102]

As described above, according to the present invention, noise shaping is performed on the spectrum of a voice signal by using the auditory weighting characteristics that are optimal for hearing, so that the noise of the reproduced signal is significantly reduced. To be improved.

[0103]

【The invention's effect】 [Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a diagram illustrating the principle of the present invention.

FIG. 3 is a diagram illustrating the principle of the present invention.

FIG. 4 is a diagram showing a voice coding transmission device as an embodiment of the present invention.

FIG. 5 is a speech coding transmission device (A according to another embodiment of the present invention.
It is a figure which shows (PC system).

FIG. 6 is a speech coding and transmitting apparatus (C according to another embodiment of the present invention.
It is a figure which shows the (ELP system).

FIG. 7 is a diagram showing a voice encoding / transmission apparatus according to still another embodiment of the present invention.

FIG. 8 is a diagram showing a speech coding and transmitting apparatus (APC system) according to still another embodiment of the present invention.

FIG. 9 is a diagram showing a voice coding and transmitting apparatus (ADPCM system) according to still another embodiment of the present invention.

FIG. 10 is a diagram showing a voice encoding and transmitting apparatus (ADPCM system) according to still another embodiment of the present invention.

FIG. 11 is a diagram showing a transmission device of a voice coding transmission device (CELP method) according to still another embodiment of the present invention.

FIG. 12 is a diagram showing a receiving device of a voice coding / transmitting device (CELP system) according to still another embodiment of the present invention.

FIG. 13 is a diagram showing a speech coding and transmitting apparatus (CELP method) according to still another embodiment of the present invention.

[Explanation of symbols]

1, 22, 40 Encoder 2, 23 Local decoder 3, 9, 11, 24, 49, 53, 427 Selection unit 4, 25, 48, 52, 437, 73, 86 Quantized noise spectrum evaluation unit 5, 30 , 46 Auditory weighting filter 6, 27 Quantizer 7, 12, 28, 31 Inverse quantizer 8, 13 Auditory weighting inverse filter 33 Prediction filter 41, 81, 91 Codebook 42, 82, 92 Gain coefficient multiplier 43, 84 _{# 1 to} 84 _{# 3} , 94 Pitch prediction filter 44, 85 _{# 1 to} 85 _{# 3} , 95 Proximity prediction filter 47 Square error minimization circuit 60, 417 Quantization error estimator 70 ADPCM encoder 71, 74 ADPCM decoder 72 _# 1 to 72 _# 3, 76 _# 1 to 76 _# 3, 78 _# 1-78 _# based on ₃ formant structure postprocessing filter 73 quantization noise spectrum evaluation unit 7 , The pitch post-processing filter based on a pitch pre-processing filter 87 _# 1 to 87 _{# 3} pitch structure based on 77 selection unit 80 CELP scheme coder 83 _# 1 to 83 _# 3, 93 _# 1-93 _{# 3} pitch structure

Claims (9)

[Claims] To 1. A transmitting side, each characteristic is different perceptual weighting filter (100 _# 1 ~100 _#n) a plurality of encoders having a (101 _# 1 ~101 _#n), provided to the encoder corresponding plurality of A plurality of transmitting side decoders (102 _{# 1} to 10 _{# 1} to 10) which decode the output signals of the corresponding encoders
2 _{# n} ) and the quantization noise generated in the corresponding encoder based on the reproduction signal from the transmitting side decoder according to a predetermined evaluation function, and the encoder with the least noise feeling is selected from the plurality of encoders. A voice code provided with an evaluation means (103) for making a judgment and a selection means (104) for selecting an output signal of an encoder with less noise selected according to the result of the judgment made by the evaluation means and transmitting it to the receiving side. Transmission device. 2. The voice coding transmission apparatus according to claim 1, wherein said transmitting side decoder is configured to perform decoding using an auditory weighting inverse filter having an inverse characteristic of the auditory weighting filter of the corresponding encoder. 3. The audio encoding / transmission apparatus according to claim 1, wherein the transmitting side decoder is included in the inside of the decoder as a part of the configuration of the corresponding encoder. 4. A transmitting side, each characteristic is different perceptual weighting filter (200 _# 1 ~200 _#n) a plurality of encoders having a (201 _# 1 ~201 _#n) and, said code device is its quantizer And a quantization noise estimation means (202) for estimating the magnitude of the perceptually weighted quantization noise generated based on the characteristics of the perceptual weighting filter, and the estimation result and input of the quantization noise estimation means. Evaluating means (20) for evaluating the noise sensation of the input speech signals of the plurality of encoders based on the speech signal according to a predetermined evaluation function to determine an encoder with less noise out of the plurality of encoders.
A voice coding transmission apparatus comprising 3) and a selecting means (204) for coding an input voice signal by using an encoder with less noise selected according to the judgment result of the evaluating means. 5. The decoder on the receiving side is provided with an auditory weighting inverse filter having an inverse characteristic of the auditory weighting filter of the encoder on the transmitting side, and the selection information indicating which auditory weighting filter is used on the transmitting side is also received. 5. The speech according to claim 1, wherein the audio signal is transmitted to the receiver side, and the decoder on the receiver side uses the auditory weighting inverse filter having the inverse characteristic of the auditory weighting filter used on the transmitter side to perform decoding. Coded transmission device. 6. The encoder on the transmission side uses a combination of a prediction filter and a perceptual weighting filter to shape the quantization noise, and the selection information indicating which perceptual weighting filter was used on the transmission side is transmitted to the reception side. The voice coding transmission device according to claim 1, wherein the voice coding transmission device is not provided. 7. A transmission side is provided with a plurality of encoders each having a perceptual weighting filter having different characteristics, the noise perceptions of the plurality of encoders are respectively evaluated, and the encoder having less noise is determined, and the noise perception is determined. A voice coding transmission apparatus configured to transmit an output signal of a low-energy encoder to a receiving side. 8. An encoder (301) for encoding an input speech signal and a plurality of perceptual weighting filters (300 _{# 1 to} 300 _#n ) having different characteristics for an output signal of the encoder on the transmitting side. A transmission side decoder (302) to be used and decoded respectively, and a noise sensation of the reproduction signal is evaluated according to a predetermined evaluation function based on the reproduction signal and the input audio signal decoded by the transmission side decoder. An evaluation means (303) for determining an auditory weighting filter with less noise feeling from the plurality of auditory weighting filters, and a selection for transmitting selection information regarding the auditory weighting filter with less noise feeling determined by the evaluation means to the receiving side. and an information transmitting means (304), to the receiving side, comprising a plurality of perceptual weighting filters same respective properties as the transmitting side is different (400 _# 1 to 400 _#n), the receiving-side apparatus , By selecting the auditory weighting filter corresponding to the selected information received from among the perceptual weighting filter of the plurality of speech coded transmission apparatus adapted to decode the received signal using the perceptual weighting filter. 9. The auditory weighting filter according to claim 1, wherein a filter based on a formant structure of a voice signal, a filter based on a pitch structure of a voice signal, or a combination of both is used. The voice coding transmission device according to any one of 1.