JPH06149298A - Speech encoding device - Google Patents

Abstract

PURPOSE:To shorten a search arithmetic processing period of time by making a search in order from a waveform pattern which has high correlativity with a waveform pattern which is selected last. CONSTITUTION:A speech signal inputted from a speech input device 1 is digitized by a digital converting device 2 by each specific quantity and compared by a comparing and selecting device 4 with waveform patterns as comparison data stored in a sound source code book memory 3. The comparing and selecting device 4 selects the waveform pattern having the highest correlativity with the speech signal, an encoding circuit 5 outputs the code corresponding to the waveform pattern, and an output device 6 sends the code to an opposite side. The comparing and selecting device 4 compares the signal with digitized data in order from the comparison data stored in a location close to the storage position of the last selected comparison data as a start point and selects the comparison data having high correlativity. Then the signal corresponding to the selected comparison data is outputted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coder.

[0002]

2. Description of the Related Art In digital voice communication, a technique has been developed in which encoded voice is vector-quantized and transmitted. In this vector quantization method, waveform patterns are stored as a codebook, the input waveform is compared with the waveform pattern, the most similar pattern is selected, and the code of the waveform pattern is converted into digital data by the other party. It is designed to be sent to. Then, the other party is configured to restore the waveform pattern based on the code. As a method for selecting a waveform pattern, a full search method and a tree search method have been developed, and FIG. 4 shows a flowchart of the full search method.

[0003]

However, the search methods described above have the drawback that the amount of calculation is large in any case. For example, in the full search method, as is apparent from FIG. 4, the initial state is returned each time a search is performed, so that there is a drawback that the search calculation amount becomes large. Therefore, the conventional speech coding device has a drawback that it is necessary to use a DPS having a large processing capacity. The present invention aims to solve the above-mentioned problems of the prior art.

[0004]

In order to achieve the above object, a speech coding apparatus of the present invention comprises means for inputting a speech signal, and a speech signal inputted by the means for inputting digital data for every predetermined amount. Means for converting to, storage means for storing a plurality of comparison data to be compared with the data digitized by the converting means, arranged and stored based on the correlation, the digitized data and the storage The comparison data stored in the means are compared with each other, the comparison data having the highest correlation with the digitized data is selected, and a position close to the starting point is the storage position of the comparison data selected last time in the selection. And means for comparing the comparison data stored in 1) with the sequentially digitized data and selecting comparison data having a high correlation.

[0005]

The input voice signal is converted into digital data for each predetermined amount. The storage means stores a plurality of comparison data to be compared with the digitized data arranged and stored based on the correlation, and the selecting means is stored in the digitized data and the storage means. The comparison data are compared with each other, and the comparison data having the highest correlation with the digitized data is selected. In making this choice,
Starting from the storage position of the comparison data selected last time, the comparison data stored in the position close to the starting point is compared with the sequentially digitized data, and the comparison data having high correlation is selected. Then, a signal corresponding to the selected comparison data is output.

[0006]

Embodiments of the present invention will be described below with reference to the drawings. In FIG. 1, a voice signal input from a voice input device 1 is digitized by a digital conversion device 2 in a predetermined amount and compared with a waveform pattern which is comparison data stored in a tone generator codebook memory 3 by a comparison / selection device 4. It is configured to be.

Waveform patterns 1 to n are stored in the tone generator codebook memory 3 as shown in FIG.
Moreover, the waveform patterns having a large correlation with each other are arranged close to each other. With reference to the waveform pattern at address k, C0, C2, C4, ... and C1, C3, C5 are located between the waveform patterns at addresses k + 1 ... k + p, k-1 ... kp.
, And have a correlation of. These correlations are C0> C2
> C4 ..., C1>C3> C5 ... Note that various arrangement modes can be adopted, such as vertically arranging in order of strong correlation with the head address as a reference. Further, in order to obtain the correlation, the autocorrelation method or the covariance method which has been widely used in the past may be used.

The comparison / selection device 4 compares the digitized voice signal with the waveform pattern of the sound source codebook memory 3, selects the waveform pattern having the highest correlation with the voice signal, and the encoding circuit 5 selects the waveform pattern. The corresponding code is output, and the code is transmitted to the other party by the output device 6.

When performing comparison, the comparison / selection device 4 first determines the correlation between the voice signal and the pattern next to the previously selected waveform pattern. Now, assuming that the address k is the previously selected waveform pattern, the next k-1 which has a high correlation with this
And k + 1 prediction residual correlation (correlation with speech signal)
Is configured to determine. And k-1 and k
The correlation of the prediction residual of +1 is compared, and the address having a large correlation is selected. If k-1 is now large, then k-1
The correlation of the prediction residual between the address and the next address k-2 is compared, and when the correlation is higher than the address k-2, the waveform pattern of the address k-1 is selected. When the correlation value of the prediction residual at the k-1 address is not higher than that at the k-2 address,
Next, the prediction residual of the waveform pattern of the address k-3, which is further adjacent to the address k-2, is calculated by correlation, and similarly compared with the pattern of the next address, and predicted from the waveform pattern of the previous address. It is configured to search for a waveform pattern having a large residual correlation value. The above configuration enables extremely efficient search. That is, the speech is characterized by a large autocorrelation, and the waveform pattern to be searched at the present time with a generally used frame length may have a large autocorrelation with the waveform pattern selected immediately before. This is because the property is high. Therefore, as described above, by comparing the correlation values of the waveform patterns on both sides of the immediately preceding waveform pattern and executing the search in the direction having the larger correlation value, the waveform pattern having the highest correlation value can be obtained in a short time. It becomes possible to search.

The operation of the comparison / selection device 4 will be described with reference to FIG. The address k of the waveform pattern selected in the immediately preceding search operation is read (step 10), and the next pattern k +
Correlation calculation of the prediction residual of the waveform patterns of 1 and k-1 is performed (steps 11 and 12). Then, the correlation values of k + 1 and k-1 are compared (step 13), when k + 1 is large, k = k + 1 is set (step 14), and the correlation value of the next k + 2 address in the + direction is calculated (step 15), k +
The correlation values of the address 2 and the preceding k = k-1, that is, the address k + 1 are compared (step 16). If the previous k + 1 address is large, this waveform pattern is selected (step 17).
If it is smaller, the correlation calculation of the next prediction residual at the address k + 3 is performed, and the correlation calculation and comparison of the addresses in the + direction are sequentially performed until a waveform pattern having a sequentially larger correlation value is searched. On the other hand, k-
When 1 is large, k = k−1 is set (step 1
8), the correlation value of the next address k−2 in the − direction is calculated (step 19), and k = k + 1 before the address k−2, that is, k
The correlation value at address -1 is compared (step 20). Previous k
If the -1 address is large, this waveform pattern is selected (step 21). If it is smaller, the correlation calculation of the next prediction residual at the address k-3 is performed, and the correlation calculation and comparison of the addresses in the-direction are sequentially performed until a waveform pattern having a sequentially larger correlation value is searched.

As described above, in the above-mentioned configuration, since the waveform pattern having the highest correlation with the previously selected waveform pattern is searched in order, the search calculation processing time can be shortened.

[0012]

As described above, the speech coding apparatus of the present invention comprises means for inputting a speech signal, means for converting the speech signal input by the means for inputting into a predetermined amount digital data. A storage unit storing a plurality of comparison data to be compared with the data digitized by the converting unit, arranged and stored based on the correlation, and the digitized data and each of the storage units stored in the storage unit. The comparison data is compared, the comparison data having the highest correlation with the digitized data is selected, and the comparison data stored at a position close to the starting point is the storage position of the comparison data selected last time in the selection. It is equipped with a means for performing sequential comparisons with the digitized data and selecting the comparison data having higher correlation than the previously selected comparison data. There is an effect of the reduction can be achieved.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an embodiment of a structure of a tone generator codebook memory 3.

FIG. 3 is a flow chart for explaining the operation of the embodiment of the present invention.

FIG. 4 is a flowchart showing an example of a conventional technique.

[Explanation of symbols]

1: voice input device, 2: digital conversion device, 3: sound source codebook memory, 4: comparison and selection device, 5: encoding circuit, 6: output device.

Claims (2)

[Claims] 1. A means for inputting a voice signal; a means for converting the voice signal input by the input means into digital data for each predetermined amount; and a comparison made with data digitized by the converting means. Multiple comparison data that should be
Storage means arranged and stored based on correlation; comparing the digitized data with each comparison data stored in the storage means, and comparing the digitized data with the comparison data having the highest correlation. When the selection is performed, the comparison data stored at a position near the starting point is compared with the digitized data sequentially from the storage position of the comparison data selected last time as a starting point, and comparison data having a high correlation is selected. And a means for outputting a predetermined signal corresponding to the selected comparison data. 2. The speech encoding apparatus according to claim 1, wherein said storage means are arranged such that the ones having the highest correlation are arranged adjacent to each other.