KR100221186B1 - Voice coding and decoding device and method thereof - Google Patents

Abstract

[purpose]

In the speech signal processing of a digital communication system, a band of an input speech signal is separated by a linear smoothing filter, and a delta modulation and a run-length encoding and a Gaussian random signal To a speech encoding and decoding apparatus and a speech encoding and decoding method capable of reducing the amount of transmission data and increasing the compression rate as well as providing a fast data processing time by encoding and decoding a speech signal.

[Configuration]

In the speech coding of the present invention, a band of an input speech signal is separated using a linear smoothing filter, the correlation between samples is increased, and then the speech signal is quantized and quantized by delta modulation and run-length coding. The quantization value, the quantization number and the amplitude level are multiplexed and transmitted. In the speech decoding of the present invention, the quantization value and the number transmitted from the speech encoding apparatus are subjected to run-length decoding to obtain the original linearity And decodes the original error signal using the amplitude level, and restores the original speech signal by summing the error signal and the voice signal passed through the linear smoothing filter.

Description

Speech coding and decoding apparatus and method thereof

Figure 1 shows a conventional delta modulation process.

FIG. 2 is a diagram showing a delta modulation process according to the present invention. FIG.

FIG. 3 is a block diagram of a speech coding apparatus according to the present invention; FIG.

FIG. 4 is a block diagram of a speech decoding apparatus according to the present invention; FIG.

FIG. 5 is a flowchart of speech coding according to the present invention. FIG.

6 is a flowchart of speech decoding according to the present invention;

DESCRIPTION OF THE REFERENCE NUMERALS

1: low-pass filter 2: subtractor

3: delta modulation unit 4: run-length encoding unit

5: amplitude level detector 6: multiplexer

11: demultiplexing unit 12: run-length decoding unit

13: delta demodulation unit 14: white noise generator

15: multiplier 16: adder

In the speech signal processing of the digital communication system, a band of an input speech signal is separated by a linear smoothing filter, and a delta modulation and a run-length encoding A speech encoding and decoding apparatus and a speech encoding and decoding method capable of reducing the amount of transmission data and increasing the compression rate as well as providing a fast data processing time by encoding and decoding a speech signal using a Gaussian random signal .

In general, a lot of information poured in a rapidly changing industrial society is transmitted through video or voice, and in particular, the information transmission method using the voice is the communication method that has been used for the longest time as a communication means or a communication means.

2. Description of the Related Art [0002] In recent years, the advantages of a digital communication method have been highlighted, and a large amount of communication has been established. Speech coders and speech decoders have been developed by digitizing and communicating voice signals.

Such a speech encoding and decoding technique requires a simple structure and a simple technology that can maintain high sound quality while requiring less data to be transmitted and suitable for real time.

Conventional methods for encoding and storing speech include a waveform coding method for storing and encoding after eliminating repetitive and unnecessary redundant components existing in a speech signal, and a method for decoding each speech signal based on a speech production model a source coding method for coding an excitation source as a filter of a excitation source and a filter element of a vocal tract by considering the source as a filter and the source coding method for the waveform coding method and the signal source coding method There is a hybrid coding method that takes advantage of this.

The waveform coding method has a disadvantage in that the naturality and intelligibility of the sound quality are very high because only the surplus components are removed in the time domain and thus the sound quality is very high. However, Examples of the encoding method include pulse code modulation (PCM), delta modulation (DM), adaptive modulation (ADM), and differential pulse code modulation (DPCM).

In addition, the speech source is generated by the sound source and the syllable filter based on the speech generation process, and the signal source coding method for linearly modeling the speech is transformed from the time domain to the frequency domain to generate excitation and formant, As such a coding method, there is an advantage that transmission memory is efficiently used by transmitting only the characteristic parameters using LPC (linear predictive coding), LSP (linear spectrum pair), PARCOR, or the like.

However, in the case of negative transition, beginning of sound, and repetition of unvoiced / unvoiced sound, deterioration of sound quality occurs because modeling can not be performed by only the characteristic parameter. Especially, all-pole model is used for modeling non- A zero-pole model is required, which is disadvantageous in that the naturalness and clarity of the sound are inferior.

In addition, the hybrid coding method using memory efficiency, which is an advantage of the high-quality sound source and the signal source coding method, which are advantages of the waveform protection method, can be classified into CELP (Code Excited Linear Prediction), VSELP Is used. In order to reduce the error of the synthesized sound, synthesis by analysis is mainly used.

In the hybrid encoding method, it is necessary to perform a repetitive calculation process and a comparison process in order to assure a high sound quality with a low bit rate. In addition, the hybrid encoding method has a disadvantage that a complexity is increased as well as a complexity and an algorithm suitable for real time is required.

In order to solve the above-mentioned disadvantages, recently, a delta modulation, which is a waveform encoding method with excellent naturalness and clarity, is used. Such a delta modulation has a higher sampling frequency and a higher bit rate than a pulse code modulation We use quantization to track the difference between the current sample and the previous sample as a 1-bit quantization value, which is widely used for sound and speech signal processing.

However, as shown in FIG. 1, as the sampling frequency increases, the delta modulation method has a problem that the amplitude of the sample changes significantly and the sample is difficult to track. Also, since the change of the encoding bit is very severe, The amount of data is enormous, and in particular, for signals mixed with noises, more data is required, so that real-time data processing becomes difficult and the restoration process at the decoding end becomes complicated.

SUMMARY OF THE INVENTION Accordingly, the present invention has been made in view of the above problems, and it is an object of the present invention to provide a method and an apparatus for separating a band of an input speech signal using a linear smoothing filter and a signal having a high correlation through a linear smoothing filter, And the error signal between the original signal and the signal that has passed through the linear smoothing filter is encoded by the amplitude level and transmitted by the quantization value and the number by the run-length encoding, and the transmitted quantization value and the number are transmitted to the run- By decoding the original signal using the amplitude level and the original error signal by decoding and delta demod, respectively, it is possible to increase the data compression rate by reducing the amount of data to be transmitted and, of course, And to provide a speech encoding and decoding apparatus and speech encoding and decoding method capable of real-time data processing.

In order to achieve the above object, the speech encoding apparatus of the present invention is a speech encoding apparatus for band-separating an input speech signal using a linear smoothing filter, enhancing a correlation between a sample and a sample for a signal passed through the linear smoothing filter 1-bit quantized value, and encoded by a quantization value and a quantization number by a subsequent run-length encoding. For the error signal between the original voice signal and the signal passed through the linear smoothing filter, the amplitude level of the Gaussian random signal The quantization value, the number of quantization levels, and the amplitude level are multiplexed and transmitted.

The speech decoding apparatus of the present invention constructs a 1-bit quantization sequence by run-length decoding the quantization value and the number transmitted from the speech coding apparatus, and then outputs the speech signal passed through the original linear smoothing filter by delta demodulation Decodes the original error signal using the transmitted amplitude level, and restores the original speech signal by summing the error signal and the voice signal that has passed through the linear smoothing filter.

Hereinafter, a speech encoding and decoding apparatus and a method thereof according to the present invention will be described in detail with reference to the accompanying drawings of FIGS. 2 to 6.

FIG. 2 is a diagram illustrating a delta modulation process according to the present invention. In the delta modulation of the present invention, a difference between a past sample and a current sample is obtained. If the difference is positive, the quantization value is '0' And '1', respectively. In contrast, the delta demodulation uses an encoded quantization value to increase the delay time when the signal is '0', and to reduce the delay when the signal is '1', thereby reconstructing the original signal.

FIG. 3 is a block diagram of a speech encoding apparatus according to the present invention. The speech encoding apparatus according to the present invention is a linear smoothing filter that performs a preprocessing process to increase a correlation between a sample and a sample by passing a speech signal having a high sampling frequency and a low quantization bit, A subtracter 2 for subtracting the original input speech signal from the signal that has passed through the low-order filter 1 and a subtractor 2 for subtracting the original input speech signal from the signal passed through the low- A run-length encoding unit 3 for delta-modulating the quantized value of the signal modulated by the delta modulation unit 3 by a delta modulating unit 3 for quantizing the quantized value, An amplitude level detector 5 for detecting an amplitude level of an error signal generated in the subtractor 2 and a quantization level detector 5 for detecting the quantization value and the quantization number of the run- ) ≪ / RTI > And a multiplexer 6 for multiplexing and transmitting bells.

In the present invention, the band of the input speech signal is separated using the low-pass filter 1, which is a linear smoothing filter, and different coding is applied according to the separated band. Particularly, The correlation between the samples becomes high, so that the 1-bit quantization value by the delta modulation facilitates tracking.

That is, since sampling is performed at a high sampling frequency, the method does not separately track changes of high frequency, so that it is easy to track samples and samples, and also has a high compression rate by reducing variation of coding bits.

The error signal between the signal having passed through the low-pass filter 1 and the original signal is analyzed by examining the average amplitude level. As a result, it has a random Gaussian characteristic with a low correlation and a large amplitude variation. By replacing the amplitude level of the signal with a Gaussian random signal, the amount of data required is reduced and the compression rate is increased and real-time processing becomes possible.

FIG. 4 is a block diagram of a speech decoding apparatus according to the present invention, which includes a demultiplexing unit 11 for demultiplexing data transmitted from a speech encoding apparatus, a demultiplexing unit 11, A run-length decoding unit 12 for performing a run-length decoding on the 1-bit quantized sequence to generate a 1-bit quantized sequence, a delta-demodulator for deleting the 1-bit quantized sequence generated by the run- A white noise generator 14 for generating a white noise and a delta demodulator 13 for decoding the amplitude level output from the demultiplexer 11 and the white noise generator 14 from the white noise generator 14 A multiplier 15 for multiplying the generated error signal by the generated white noise, an error signal output from the multiplier 15 and a signal output from the delta demodulator 13 to output an original speech signal, And an adder 16 for adding an adder.

Next, encoding and decoding of the speech signal according to the present invention will be described with reference to FIGS. 5 and 6. First, as shown in the flowchart of FIG. 5, the speech signal is encoded with a high sampling frequency and a quantization level In order to encode the 1-bit input voice signal, the input voice signal is passed through the low-pass filter 1 (S1).

That is, the above process is a linear smoothing filter process for increasing the correlation between the sample and the sample,

Where S (.) Is the input speech digital signal and k is the delay factor.

Then, it is determined whether the signal filtered in the step S1 is an error signal between the signal passing through the low-pass filter 1 and the original voice signal. If the signal passed through the low-pass filter 1 is a delta And modulated by the modulation unit 3 (S3).

At this time, the delta modulation obtains the difference between the past sample and the current sample, and assigns a quantization value to each of '0' if the difference is positive and '1' if it is negative to construct a bit string.

Subsequently, after the delta modulation in the step S3, the run-length encoding unit 4 performs run-length encoding (S4). At this time, in order to optimize the quantization value, The current quantization value is compared with the current quantization value, and if the quantization values are the same, the number of quantization values is counted, and if the quantization values are different, the changed quantization values are compared again, and the quantization number is counted.

For example, if the delta-modulated quantization value is '111111000 0011111', the quantization value of the first '1' is 6, the quantization value of the second '0' is 5, and the quantization value of the third '1' is 5 Dogs.

Accordingly, the quantization value and the quantization number are transmitted at the time of transmission by the run-length encoding as described above.

On the other hand, if it is determined in step S2 that the signal of step S1 is an error signal between the signal passed through the low-pass filter 1 and the original speech signal, the amplitude level of the error signal is (S6) and transmits its amplitude level as a Gaussian random signal.

At this time, the amplitude level M is obtained by Equation 2 and Equation 3 below.

Then, the quantization value, the number of quantization and the amplitude level are multiplexed and transmitted through the multiplexing unit 6 (S6). As a result, the input speech digital signal is encoded and transmitted with the quantization value, the number of quantization, and the amplitude level.

Next, description will be made on the decoding of the speech signal according to the present invention with reference to the flowchart of FIG. 6. First, the data transmitted from the speech coding apparatus is demultiplexed by the demultiplexing unit 11 into a quantization value, a quantization number and an amplitude level S11), it is determined whether the demultiplexed signal is a quantized value and a quantized number (S12). If the quantized value and the quantized number are the quantized value and the number of quantized values, the run-length decoding unit 12 reads the quantized value and the number, To form a 1-bit quantization column (S13).

Next, the delta demodulation unit 13 performs delta demodulation on the 1-bit quantization sequence constructed in the step S13 to recover the original signal passed through the low-pass filter 1 (S14).

That is, the run-length decoding and the delta demodulation process reads the quantization value and increases after the delay if the quantization value is "0". If the quantization value is "1", the run-length decoding and the delta demodulation decrease the delay and decode the original signal.

For example, if the quantization value and its number are "160515 ", the original signal is reconstructed by decreasing 6 times, then incrementing 5 times, and then decrementing 5 times.

Therefore, if S _L (n) = 0, S _L (n + 1) = -1, S _L (n + 2) = -2, ... , S _L (n + 5) = -5, S _L (n + 6) = -4, S _L (n + 7) = -3, ... , S _L (n + 10) = -1, and S _L (n + 11) = -2 and S _L (n + 12) = -3. , And S _L (n + 15) = - 6.

If the signal demultiplexed in step S12 is an amplitude level, the transmitted amplitude level M and the white noise generated in the white noise generator 14, that is, the Gaussian random signal G (.), (15) to restore the original error signal (S15).

If the process at this time is expressed by Equation 4,

Finally, the original signal passed through the low-pass filter 1 restored in the step S14 by the adder 16 is added to the error signal restored in the step S15 to restore the original speech digital signal (S16).

As described above, according to the present invention, unnecessary high-frequency components are replaced with Gaussian components in performing encoding and decoding according to separate bands, thereby reducing the amount of data required. In particular, , Which has a very high compression ratio due to linear smoothing.

Claims (2)

An apparatus for encoding and decoding speech signals in a digital communication system, the apparatus comprising: a low-pass filter (1) for performing a preprocessing process for passing an input speech signal to increase a correlation between a sample and a sample; A subtracter 2 for subtracting the original input speech signal from the passed signal to generate an error signal, a delta modulator 3 for delta modulating the signal passed through the low-pass filter 1, A run-length encoding unit 4 for encoding the quantized value of the modulated signal and the number of quantized values of the signal modulated by the quantization unit 3 and encoding the quantized value and the number of quantized values, And a multiplexer 6 for multiplexing and transmitting the quantization values and quantization numbers of the run-length encoding unit 4 and the amplitude levels detected by the amplitude level detecting unit 5, The encoding device ; A demultiplexer 11 for demultiplexing the data transmitted from the speech encoding apparatus, a demultiplexer 11 for decoding the quantized values and the quantization numbers output from the run-length decoding unit 12, A delta demodulator 13 for delta-demodulating the signal after the decoding by the decoding unit 12 to decode the signal passed through the original low-pass filter 1, a white noise generator 14 for generating white noise, A multiplier 15 for multiplying the amplitude level output from the multiplier 15 by the white noise generated by the white noise generator 14 to decode the original error signal; And an adder (16) for adding the decoded signal from the demodulator (13) and outputting the original input speech signal. ≪ Desc / Clms Page number 13 > A method for encoding and decoding a speech signal in a digital communication system, the method comprising: a first step of performing low-pass filtering of an input speech signal in order to band-separate the speech signal; A third step of delta-modulating the speech signal passed through in the first step; a third step of performing a run-length encoding of the delta-modulated signal by the quantization value and the quantized number in the third step; A fifth step of detecting an amplitude level of the error signal generated in the second step, a quantization step of quantizing the run-length-encoded quantized value and the number in the fourth step and the amplitude level detected in the fifth step, A speech coding step of performing a speech coding process in a sixth step; A seventh step of demultiplexing a signal transmitted by the speech encoding process and outputting a quantization value and a number and an amplitude level; and a seventh step of performing a run-length decoding of the quantization value and the number output from the seventh step, A ninth step of delta-demodulating the quantized sequence decoded in the eighth step into an original low-pass filtered speech signal, and an eighth step of restoring the original error signal by multiplying the amplitude level output from the seventh step by the white noise And an input speech signal reconstructing step of reconstructing an original input speech signal by adding the low-pass filtered speech signal delta-demodulated in the ninth step and the error signal reconstructed in the tenth step, And decoding the encoded audio data.