KR100348137B1 - Voice coding and decoding methods by using sampling rate conversion - Google Patents

Abstract

PURPOSE: Voice coding and decoding methods by using a sampling rate conversion are provided to improve the coding efficiency by reducing a dynamic range of a residual signal obtained by linear-predicting and analyzing a voice signal. CONSTITUTION: A first residual signal is generated by linear-predicting and analyzing a frame unit of a voice signal(S102). The first residual signal generated by the first linear predicting analyzing process is interpolated(S104). A second residual signal is obtained by linear-predicting and analyzing the first residual signal interpolated in the interpolating process to a lower degree than a degree used in the first linear predicting analyzing process(S106). The second residual signal is decimated(S108). The residual signal decimated through the decimation process is coded(S110).

Description

Speech coding and decoding method by sampling rate conversion

The present invention relates to a method for encoding a speech signal, and more particularly, to an encoding method for enabling efficient encoding by reducing a dynamic range of a cold order signal obtained by linear predictive analysis. It relates to a decoding method.

The voice encoding method that reduces the amount of information by removing the redundancy of the voice signal increases the transmission efficiency when transmitting the voice signal and reduces the storage capacity when storing the voice information. The speech encoding method can be broadly classified into a waveform encoding method, a sound source encoding method, and a hybrid encoding method in which the two methods are mixed.

In the hybrid coding method, formant information is normally encoded by linear prediction encoding, and residual exccited linear prediction (RELP) and vector-excited linear prediction (VSELP) are determined according to how the residual signal is encoded. Vector Sum Excited Linear Prediction (MPLP), Multipulse-Excited Linear Prediction (MPLP), and Code Excited Linear Prediction (CELP) have been proposed.

In general, a speech signal is modeled as a linear prediction filter that assumes a quasi-periodic signal and inputs a pulse train or white noise. That is, it is assumed that the voice s (n) is predictable from the previous p signals {s (n-1), s (n-2) ,,, s (np)}.

Expressed as a formula, the prediction signal of s (n) Is expressed as

Where {α _i } is linear predictive coefficients. If the difference between the original signal and the prediction signal (residual signal) is e (n)

Is given by

e (n) can be modeled in various ways, but is mainly used as a pulse train when voice is voiced and white noise when voice is used as an excitation signal of a filter.

In the conventional speech analysis and the speech encoding apparatus using the same, a certain section of the speech is modeled using a 10th order LPC filter.

However, in this modeling method, since the speech cannot be accurately modeled, the dynamic range of the residual signal increases. Therefore, there is a problem that the residual signal e (n) must be represented by a large number of bits in order to ensure the high quality of the synthesized speech.

It is an object of the present invention to provide an improved encoding method which can reduce the number of bits representing a residual signal by reducing the dynamic range of the residual signal.

Another object of the present invention is to provide a decoding method corresponding to the above encoding method.

The encoding method according to the present invention to achieve the above object

In the speech encoding method of linearly predicting and analyzing a speech signal in a frame unit to obtain a residual signal, and encoding the residual signal,

A first linear prediction analysis process of generating a first residual signal by linearly predicting and analyzing a speech signal in a frame unit;

Interpolating a first residual signal generated in the first linear prediction analysis process;

A second linear prediction analysis process of linearly analyzing the first residual signal interpolated in the interpolation process to an order lower than the order used in the first linear prediction analysis process to obtain a second residual signal;

Reducing the second residual signal: and

And encoding the residual signal reduced by the reduction process.

The onset reduces the dynamic range of the residual signal through iterative interpolation and reduction, enabling accurate modeling of speech and the transmission of these parameters with fewer bits. Therefore, it is possible to effectively implement a low rate speech coder.

The decoding method according to the present invention to achieve the above another object

In a method of restoring a speech signal from a transmitted parameter by performing linear predictive analysis by interpolating a first residual signal obtained by linear prediction encoding a speech signal, and then reducing and encoding a second residual signal obtained therefrom,

Restoring the second residual signal from the transmission parameter;

Interpolating the second residual signal restored in the restoration process;

A first synthesis process of synthesizing the second residual signal interpolated through the interpolation process through a synthesis filter;

A reduction process of generating a first residual signal by reducing the residual signal synthesized through the first synthesis process; And

And a second synthesis process of synthesizing the first residual signal obtained through the reduction process through a synthesis filter. Hereinafter, with reference to the accompanying drawings will be described in detail the features and effects of the present invention.

1 is a flowchart showing an encoding method according to the present invention.

First, in the first linear prediction analysis process (S102), a first residual signal is generated by linearly predicting and analyzing a speech signal in a frame unit.

In the interpolation process S104, the first residual signal generated in the first linear prediction analysis process S102 is interpolated.

In the second linear prediction analysis process S106, the second residual signal is obtained by linearly predicting and analyzing the first residual signal interpolated in the interpolation process S104 at a lower order than the order used in the first linear prediction analysis process S102. .

In the reduction process S108, the second residual signal obtained in the second linear prediction analysis process S106 is reduced.

In the encoding process S110, the residual signal reduced through the reduction process S108 is encoded.

2 is a flowchart showing a preferred embodiment of the encoding process shown in FIG. In the embodiment illustrated in FIG. 2, the first linear prediction analysis process S102 of FIG. 1 is performed in step 202 (S202), and the interpolation process S104 of FIG. 1 is performed in step 212 (S112) in FIG. 1. The two-linear prediction analysis process S106 corresponds to step 214 (S214), the reduction process S108 of FIG. 1 corresponds to step 216 (S216), and the encoding process S110 of FIG. 1 corresponds to step 218 (S218). do.

In step 200 (S200) to step 202 (S202), linear prediction coefficients (hereinafter referred to as LPC coefficients) are obtained from the input frame-based speech, and a residual signal obtained by removing the prediction signal from the original audio signal is obtained using the same.

The input speech signal s (n) is modeled by p-order linear prediction analysis.

Then, let m = 0 and do the following.

The residual signal is obtained from the modeled speech signal as follows.

Here, e _m (n) is a residual signal at the m-th iteration, α _mi is a linear prediction coefficient at the m-th iteration, and s (ni) are speech signals of the previous frame. The second term of Equation 4 represents the predicted speech signal for the speech signal of the current frame.

In step 204 (S204) it is determined whether the repetition number m is less than the given maximum repetition number Mmax. If the number of repetitions m is not less than the given maximum number of repetitions Mmax, the process branches to step 218 (S218), and if greater, the process proceeds to step 206 (S206).

In step 206 (S206) it is determined whether the number of pulses is less than the threshold. If the number of pulses is less than the threshold, the process branches to step 218 (S218); otherwise, to step 207 (S207).

Here, if the length of the frame is N, the number of pulses can be defined by the following equation.

Here, δ ₁ and δ ₂ are predetermined thresholds.

In step 207 (S207), the power P _e of the residual signal is calculated. Power P _e is calculated by the following equation.

In step 208 (S208) and determines whether the P _e is smaller than a predetermined threshold value, if 0.208 in step (S208) P _e is smaller than a predetermined threshold value (P) branches to step 218 (S218).

Where P is a predetermined threshold.

On the other hand, in step 208 (S208) if P _e is greater than or equal to the predetermined threshold value P proceeds to step 210 (S210).

In step 210, the number of repetitions m is increased.

In step 212, interpolation is performed on the residual signal. Interpolation is performed to add redundancy to the residual signal and experimentally the value of interpolation factor I should be at least 5.

The interpolated residual signal is obtained by the following formula {e ¹ _m (n), n = 0 ,,, NI -1}.

In step 214 (S214), a linear prediction analysis is performed on the interpolated residual signal, and the residual signal is obtained again based on the obtained LPC coefficients. In this case, the order of the linear predictive analysis uses p _l less than the analysis order p used in step 202 (S202).

Here, the total number of samples of the residual signal is NI, which is the product of the interpolation factor I and the frame length N.

The residual signal e ^I _m (n) is obtained as follows. Here, the subscript I means the number of repetitions.

In step 216 (S216), the reverse process of step 212 (S212) is reduced. The reduction factor to be used is I equal to the interpolation factor.

The residual signal after the reduction can be expressed as {e _m (n), n = 0 ,,, N -1}.

On the other hand, in step 218 (S218) is modeled for the residual signal of the m order.

The modeling method uses a codebook method to minimize the number of bits. Therefore, the transmission parameters include the iteration number m, the codebook index of the mth order residual signal, and the m LPC coefficients. Becomes

In practice, however, there is no need to transmit anything except the initial LPC coefficient. At this time, the LPC coefficients obtained for the arbitrary speech section are commonly used in the encoding and decoding of the speech.

3 is a flowchart illustrating a method of decoding a speech using parameters transmitted by the encoding method illustrated in FIG. 1.

In the decoding method according to the present invention as shown in FIG. 3, first, in the reconstruction process (S302), the second residual signal is reconstructed from the transmission parameter.

In the interpolation process S304, the second residual signal reconstructed in the reconstruction process S302 is interpolated.

In the first synthesis process S306, the second residual signal interpolated through the interpolation process S304 is synthesized through a synthesis filter.

In the reduction process S308, the synthesized residual signal is reduced by performing the first synthesis process S306 to generate a first residual signal.

In the second synthesis process S310, the first residual signal obtained through the reduction process S308 is synthesized through a synthesis filter.

4 is a flowchart showing a preferred embodiment of the decoding method shown in FIG. In the decoding method shown in FIG. 2, the reconstruction process S302 of FIG. 3 is performed in step 402 (S402), the interpolation process of FIG. 3 (S304 in step 406 (S406), and the first synthesis process of FIG. S306 corresponds to step 408 (S408), the reduction process S308 of FIG. 3 corresponds to step 410 (S410), and the second synthesis process S310 of FIG. 3 corresponds to step 414 (S414).

In operation 400, the input transmission parameter is input.

In step 402, the m-th order residual signal has a codebook index extracted from the transmission parameter. Restore

In step S404, it is determined whether m extracted from the transmission parameter is 0. If m is 0, the flow branches to step 414. Otherwise, the flow proceeds to step 406 (S406).

In step 406, the residual signal reconstructed by the interpolation factor I used in the encoding process illustrated in FIG. Interpolate

In step 408, the interpolated residual signal is synthesized using a synthesis filter having an LPC coefficient {α _m1 ,,, α _mp ).

In operation 410, the synthesized signal is reduced to the reduction factor I used in the encoding process illustrated in FIG. 1. This signal is the residual signal of (m-1) Becomes

This process is repeated until m = 0.

Step S414 (S414) is the final residual signal Is excited by a synthesis filter having LPC coefficients obtained in the encoding process shown in FIG. 1 to reproduce speech.

As described above, the speech coding method according to the present invention enables accurate modeling of speech by reducing the dynamic range of the residual signal through iterative interpolation and reduction, and transmits these parameters with a small number of bits. Make it possible. Therefore, it is possible to effectively implement a low rate speech coder.

The speech coding method according to the present invention has a very small number of transmission parameters, so that a voice coder of 4 kbps or less can be realized. It is possible.

1 is a flowchart showing an encoding method according to the present invention.

2 is a flowchart showing a preferred embodiment of the encoding process shown in FIG.

3 is a flowchart illustrating a method of decoding a speech using parameters transmitted by the encoding method illustrated in FIG. 1.

4 is a flowchart showing a preferred embodiment of the decoding method shown in FIG.

Claims (13)

In the speech encoding method of linearly predicting and analyzing a speech signal in a frame unit to obtain a residual signal, and encoding the residual signal, A first linear prediction analysis process of generating a first residual signal by linearly predicting and analyzing a speech signal in a frame unit; Interpolating a first residual signal generated in the first linear prediction analysis process; A second linear prediction analysis process of linearly analyzing the first residual signal interpolated in the interpolation process to an order lower than the order used in the first linear prediction analysis process to obtain a second residual signal; Reducing the second residual signal; And And encoding the residual signal reduced by the reduction process. The method of claim 1, And repeating the interpolation process, the second linear prediction analysis process, and the reduction process until the result of the first linear prediction analysis process satisfies a condition in which the number of pulses of the residual signal is less than a predetermined threshold. The method of claim 1, wherein the interpolation process, the second linear prediction analysis process, and the reduction process are repeated until the result of the first linear prediction analysis process satisfies a condition that the power of the residual signal is less than a predetermined threshold. Speech coding method. The speech encoding method of claim 1, wherein the interpolation factor of the interpolation process and the reduction factor of the reduction process are the same. The method of claim 1, The interpolation factor of the interpolation process and the reduction factor of the reduction process are at least five or more. The speech encoding method of claim 1, wherein an analysis order of the second linear prediction analysis process is less than an analysis order of the first linear prediction analysis process. The speech encoding method of claim 1, wherein the encoding process is performed using a codebook. The speech encoding method of claim 1, wherein the encoding process encodes a linear prediction coefficient resulting from the first-order linear prediction analysis process. In a method of restoring a speech signal from a transmitted parameter by performing linear predictive analysis by interpolating a first residual signal obtained by linear prediction encoding a speech signal, and then reducing and encoding a second residual signal obtained therefrom, Restoring the second residual signal from the transmission parameter; Interpolating the second residual signal restored in the restoration process; A first synthesis process of synthesizing the second residual signal interpolated through the interpolation process through a synthesis filter; A reduction process of generating a first residual signal by reducing the residual signal synthesized through the first synthesis process; And And a second synthesis process of synthesizing the first residual signal obtained through the reduction process through a synthesis filter. The method of claim 1, wherein the residual signal e _m (n) is derived from the speech signal modeled in the first linear prediction analysis. Is, Where s (n) is the input speech signal modeled by p-order linear prediction analysis, The speech coding method characterized by the above-mentioned. 2. The interpolated residual signal of claim 1, wherein {e ¹ _m (n), n = 0 ,,, NI -1} The speech coding method characterized by the above-mentioned. The method of claim 1, wherein the residual signal e ^I _m (n) in the second linear prediction analysis process is (Where I is an interpolation factor). The method of claim 1, wherein the residual signal after reduction in the reduction process is {e _m (n), n = 0 ,,,, N −1}. e _m (n) = e ^I _m-1 (nI) n = 0 ,, N −1.