JPH04291400A - Low delay code device type predictive encoding method - Google Patents

Abstract

PURPOSE:To improve the quantity of an encoded voice. CONSTITUTION:A quantized amplitude is selected and supplied for a pitch period candidate selected by a pitch excitation source, a synthesizing filter is driven with the pitch signal to synthesize a voice, and a filter coefficient is linearly predicted from a past decoded voice waveform and set in the synthesizing filter; and the pitch period candidate and a quantization amplitude value are selected so that the distortion of the synthesized voice to an input voice becomes minimum, thus encoding the voice. The autocorrelation coefficient of the past decoded voice is calculated and either of two groups of four quantization gains g0-g3, and g0'-g3' are selected according to whether the calculated coefficient is >=0.6 or not to determine the amplitude of the pitch period candidate by adapting this method.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] This invention selects a pitch candidate from a pitch excitation source, gives a quantized amplitude to the pitch candidate, drives a synthesis filter, and uses the filter coefficients of the synthesis filter to generate the previously decoded audio. Low-delay code drive that linearly predicts and sets the waveform, uses the synthesis filter to determine pitch candidates and quantization amplitude values, encodes and outputs in frames with a short number of samples, and encodes audio with less delay. This invention relates to a predictive coding method.

0002

2. Description of the Related Art In fields such as digital mobile communications, various high-efficiency encoding methods are used to effectively utilize radio waves. Methods for encoding with an amount of information of about 8 kbit/s include CELP (code-driven linear prediction);
VSELP (vector addition driven linear prediction), multipulse coding, etc. are known.

In these systems, as shown in FIG. 3, a filter coefficient determination unit 11 calculates a prediction coefficient from a plurality of samples of input audio to determine a filter coefficient, and sets the filter coefficient in a synthesis filter 12. A(Z) is a transfer function of the synthesis filter 11. The pitch period extracted from a plurality of pitch period components (excitation candidates) of the pitch excitation source 13 and the candidates extracted from a plurality of noise waveform vectors (for example, random number vectors, excitation candidates) of the codebook excitation source 14 are each acquired by a gain unit 15. , 16, and then add the quantized gain (amplitude) and supply it as a driving signal to the synthesis filter 12 to synthesize speech, and then calculate the power so that the distortion of the synthesized speech with respect to the input speech is minimized. In the section 17, both excitation sources 13,
14 are selected, and the gains of gain sections 15 and 16 are set. The code output unit 18 outputs the code numbers and gains selected for each of the prediction coefficients, pitch period component candidates, and codebook candidates as codes.

[0004] Prediction coefficients for determining filter coefficients of the synthesis filter 12 are obtained by analyzing input speech. 20-3
Processing is performed with approximately 0 ms (usually 128 or 256 samples) as one frame. In the forward prediction type in which prediction coefficients are obtained from one frame before the sample to be encoded, the encoded output is delayed by at least one frame. In these methods, since one frame is long, a large delay occurs.

[0005]Currently, in applications such as personal communications, there is a demand for voice encoding methods with less delay, and methods that cause large delays such as those described above are not desirable. As a low-latency audio encoding method, 16kbi
LD-CELP (low delay code-driven linear prediction) at t/s
Encoding methods are known. This method uses backward prediction type pitch prediction and proximity prediction. That is, instead of using the signal in the frame that is currently being quantized to calculate the prediction coefficient, the encoded output is stored in the storage/decoding unit 19 as shown by the broken line in FIG. 3, and this past code is decoded. ,
A filter coefficient determination unit 11 windows the decoded speech and performs linear prediction including pitch periodicity via a correlation function. In other words, the waveform of the past frame is decoded, the filter coefficient of the synthesis filter 12 is determined from the waveform, and the synthesis filter 1/A0 (Z) is used to determine the pitch parameter candidate in the pitch excitation source 13 and its amplitude ( In addition to determining the quantized value (gain), the shape vector (noise component) candidate in the codebook excitation source 14 and its amplitude quantized value are determined, and the codes of both of the determined candidates and the quantized value are transmitted.

In this method, previously decoded speech can be commonly used by both the encoder and the decoder, so there is no need to transmit information on prediction coefficients or periodicity (pitch). Therefore, the number of samples per frame should be small, e.g.
The number of samples can be set to 10, the frame length can be shortened, and encoding with little delay is realized.

However, since LD-CELP predicts the current frame only from past decoded sequences, the prediction error is larger than in the conventional forward prediction type. Therefore,
When encoding at approximately 8 kbit/s, waveform distortion increases rapidly and quality deteriorates. In order to achieve low-latency speech encoding with an information amount of about 8 kbit/s, the pitch periodicity is not included in linear prediction as in LD-CELP, but the pitch period component is also extracted from the decoded speech. A method has been proposed.

However, in all of the conventional methods, the quantization width of the pitch gain (amplitude) is a constant value that is appropriately determined from the maximum dynamic range depending on the desired encoding accuracy, and the quantization width of the pitch gain (amplitude) is set to a constant value that is appropriately determined from the maximum dynamic range depending on the desired encoding accuracy. No scope was given. That is, when searching for pitch period component candidates from the pitch excitation source 13, g and C are determined that minimize the distortion d to the input voice expressed by the following equation.

d=(X-g・H・C)2 X; input speech g; pitch gain H; impulse response C of synthesis filter 12; pitch component candidate It is given by the following formula.

g=(X,C)/|C|2 The distribution of the optimal gain of the pitch period component for the correlation coefficient in each frame obtained by analyzing approximately 1000 seconds of 100 sentences is as shown in FIG. Became. In FIG. 4, for example, if the correlation coefficient is between 0.9 and 1.0, the optimal gain is between 0.9 and 1.0.
This shows that there were 1,560 frames. As can be understood from FIG. 4, the distribution of the optimal gain (amplitude) differs depending on the value of the correlation coefficient, and the larger the correlation coefficient, the narrower the width of the optimal gain. However, in the past, this relationship was not known and a constant quantization width was used, which did not necessarily result in good encoding.

[0011] An object of the present invention is to perform speech encoding with low delay based on information on past encoded speech.
By adaptively controlling the pitch gain quantization width,
An object of the present invention is to provide a low-delay code-driven predictive encoding method that can realize high-quality speech encoding.

0012

[Means for Solving the Problems] According to the present invention, in a low-delay code-driven predictive coding method, the autocorrelation of past decoded speech is determined, and the pitch gain ( Adaptively controls the quantization width of (amplitude). This control parameter is
〇 Maximum value of autocorrelation function 〇 Correlation coefficient at the period selected in conditional pitch prediction can be considered. Furthermore, in the latter case, select multiple candidates for periods with large autocorrelation, find the most desirable period for pitch prediction of the current frame from among them, and use the correlation coefficient corresponding to that period as a parameter to determine the gain quantization width. Method 1 to decide.

[0013] Select a plurality of candidates with periods with large autocorrelation, determine the quantization width for each candidate based on the corresponding autocorrelation value, and pair the most desirable period and gain for pitch prediction of the current frame. Method 2: Method 3: A quantization width is determined based on the correlation coefficient in the period used in the previous pitch prediction, and the quantization width is commonly used for all period candidates to quantize the current pitch gain. and so on.

[0014]

[Embodiment] According to the present invention, in the low-delay code-driven predictive coding method as described above, the autocorrelation of past decoded speech is determined, and the pitch gain in the current frame is calculated using the correlation function as a parameter. Adaptively control the quantization width. In other words, as shown in FIG. 4, when the correlation is high, the optimum gain is concentrated around 1, and as the correlation becomes low, the range of the optimum gain value becomes wider.

Therefore, in this example, taking advantage of the high correlation between adjacent frames, the value of the correlation coefficient of the candidate selected in the previous frame is used, and as shown in FIG. Depending on whether the correlation coefficient value is 0.6 or more or 0.6 or less, the pitch gain can be set to two sets of 4 states g0 to g3 or g0'
~g3', that is, quantize with 2 bits. The selection of one of these two sets of four states is based on the autocorrelation of past decoded speech, where the previous frame has a value of 0.6.
Since the decision is made based on whether or not the above is true, the receiving side can also make the decision, and the pitch gain can be quantized using 2 bits.

In the above, the quantization width was adaptively changed depending on whether the correlation coefficient value was 0.6 or more or less, but as shown in FIG. Good too. The example in FIG. 2 is a case of 8 divisions, which allows quantization closer to the optimal gain distribution. In the above two examples, the value of the previous frame is used as the correlation coefficient value of the current frame. This is predicted by performing linear prediction using a series of correlation coefficient values of candidates selected in the past. This allows pitch gain to be quantized with a value close to the value of the correlation coefficient of the candidate that will be selected in the current frame.

[0017]

[Effects of the Invention] As described above, according to the present invention, the quality of encoded speech is improved by adaptively controlling the quantization width of the pitch gain in the current frame based on the information of past decoded speech. can be improved.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing a numerical example when the quantization width of pitch gain is changed depending on whether the correlation coefficient is 0.6 or more or less than 0.6 in an embodiment of the present invention.

[Fig. 2] The value of the correlation coefficient in the embodiment of this invention is 10.
The figure which shows the numerical example when the quantization width is changed depending on which of the divisions it is.

FIG. 3 is a block diagram showing a general configuration of a low-delay code-driven predictive encoding method.

FIG. 4 is a diagram showing the relationship between the value of an autocorrelation coefficient and the optimum pitch gain.

Claims (1)

[Claims] Claim 1: Select a pitch candidate from a pitch excitation source using a relatively short number of samples of an audio signal as one frame, apply a quantized amplitude to the pitch candidate, drive a synthesis filter, and apply the quantized amplitude to the pitch candidate. The filter coefficients are set by linear prediction from the audio waveform decoded up to one frame before, and the synthesized filter is used to determine the pitch candidate and the quantization value, and the predictive code is encoded in units of one frame. In the encoding method, the correlation function of the audio waveform decoded up to one frame before is obtained, and the quantization width of the pitch amplitude used for encoding the current frame is adaptively controlled according to the value of the correlation function. A low-delay code-driven predictive coding method.