JPH0650438B2 - Speech coder - Google Patents

Description

The present invention relates to a speech coder used for speech band compression, speech accumulation, and the like.

〔Overview〕

The present invention, in a multi-pulse drive type speech coder that encodes a prediction residual signal with a plurality of pulse examples, generates an influence signal from a frame immediately before a frame currently being analyzed, and uses this influence signal from a weighted signal. By subtracting the signal and obtaining the cross-correlation function with the impulse response function, the quality deterioration due to the influence signal from the immediately preceding frame is eliminated.

[Conventional technology]

With the recent development and diversification of data networks, voice band compression technology and voice storage technology are required to have a low bit rate for the purpose of reducing line cost and network efficiency.

Conventionally, AD has been used as a coding method in the vicinity of 16 bit / s of voice.
-PCM, APC, ATC, etc. have been proposed (for example, Ochiai, Araseki, "Speech Coding Technology", Journal of Information Processing Society of Japan, Vol.24, No.8 PP.976-893, 1983), but recently predicted A multi-pulse drive type speech encoding method for encoding a residual signal with a plurality of pulse trains has been announced, and it is promising because of its quality / bit rate ratio (for example, Araseki, “Recently using a source generation model Speech Encoding Technology ", IEICE, Information Theory Workshop Material, IT84-31, 1984, Ono, Araseki," Study on Multi-pulse Driven Speech Encoding Method ", IEICE, Communication Method Workshop Material, CS82 -161, 1982).

This method is suitable for voice coding of 8 to 16 kbit / s, and meets the above-mentioned requirements in the field of voice band compression and voice storage.

[Problems to be solved by the invention]

However, in the conventional multi-pulse drive type encoding method, a problem when actually constructing the encoder is the influence of the boundary between the adjacent speech analysis frames. Since a voice signal is originally continuous, when a certain voice analysis frame is focused, the voice analysis frame has a high correlation with the previous voice analysis frame. In the case of multi-pulse driven speech coding, if a pulse was present at the end of the previous speech analysis frame, most of the impulse response of that pulse is in the current speech analysis frame. Therefore, if pulse extraction is performed by focusing only on the current voice analysis frame, it means that the pulse component of the previous voice analysis frame is also included in it, and the duplicated pulse becomes an abnormal sound, and the reproduced sound quality is improved. There is a drawback that it deteriorates.

It is an object of the present invention to provide a higher quality multi-pulse driven speech coder by eliminating the above drawbacks.

[Means for solving problems]

The present invention provides a spectrum analyzer for obtaining a spectral coefficient and an impulse response function in an analysis frame of an input speech signal, a weighting filter for weighting the input speech signal with the spectral coefficient to generate a weighted signal, and the impulse. A cross-correlator that obtains a cross-correlation function between the response function and the weighted signal, an auto-correlator that obtains an auto-correlation function of the impulse response function, and a pulse train is extracted and output from the cross-correlation function and the auto-correlation function. In a speech coder including a pulse extraction circuit, a weighting signal correction circuit for generating an influence signal from an analysis frame immediately before an analysis frame currently being analyzed from the pulse train and the spectrum coefficient, and the influence signal from the weighted signal. Generate a corrected weighting signal from which And a subtractor for inputting the input power.

[Action]

The present invention, in the pulse train of the frame immediately before the frame currently being analyzed, which is output from the pulse extraction circuit by the weighting correction circuit, generates an influence signal from the pulse that affects the frame currently being analyzed, and by the subtracter, A corrected weighting signal is generated by subtracting this influence signal from the weighting signal output from the weighting filter, is input to the cross-correlator, and the cross-correlation function with the impulse response function is obtained.

Therefore, abnormal noise is not generated due to the influence signal of the immediately preceding frame, and the quality can be improved.

〔Example〕

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. In this embodiment, a spectrum analyzer 2 for obtaining a spectral coefficient and an impulse response function in an analysis frame of an input speech signal, and a weighting filter 3 for weighting the input speech signal with the spectral coefficient to generate a weighted signal. , A cross-correlator 7 for obtaining a cross-correlation function of the impulse response function and the weighted signal, and an auto-correlator 6 for obtaining an auto-correlation function of the impulse response function.
And a pulse extraction circuit 8 that extracts and outputs a pulse train from the cross-correlation function and the autocorrelation function, and generates an influence signal from the analysis frame immediately before the analysis frame currently being analyzed from the pulse train and the spectral coefficient. It includes a weighting signal correction circuit 4 and a subtractor 5 for generating a correction weighting signal by subtracting the influence signal from the weighting signal and inputting it to the cross-correlator 7. In the figure, 1 is an input terminal, 9 is a pulse output terminal, and 10 is a spectrum coefficient output terminal.

The feature of the present invention resides in that the weighting signal corrector 4 and the subtractor 5 are provided in FIG.

Next, the operation of this embodiment will be described with reference to the waveform chart of each part shown in FIG. Now, input terminal 1 is shown in FIG.
It is assumed that the input voice signal 11 having the voice waveform shown in (a) is input. As shown in the figure, the analysis length is N samples (for example, 20 ms), the pulse extraction frame length is M samples (for example, 3 ms), and the zero time is the position shown in the figure.
The current analysis frame is at time l, the analysis frame immediately before is at time (l-1), and the analysis frame immediately after is at time (l + 1) frame. The input audio signal 11 is input to the spectrum analyzer 2 and the weighting filter 3. In the spectrum analyzer 2, the spectrum coefficient 12 in the analysis frame and the second
The impulse response function 13 shown in FIG. 6 (b) is obtained. The spectrum coefficient 12 is output to the weighting filter 3, the weighting signal corrector 4, and the spectrum coefficient output terminal 10, and the impulse response function 13 is used for the autocorrelator 6 and the crosscorrelator 7.
Is output to. Here, the impulse response function 13 shown in FIG. 2 (b) is an impulse response function of M sample length based on the spectral coefficient (for example, PARCOR coefficient) analyzed from the frame at time l. The weighting filter 3 weights (N + M) samples of the input input audio signal 11 with the spectral coefficient 12 and outputs the second weighted signal.
The weighting signal 14 having the waveform shown in FIG. In the weighting signal corrector 4, in order to remove the influence of the pulse due to the analysis of the (l-1) time, the signal is fed back from the pulse extractor 8 shown in FIG. 2 (d) analyzed at the (l-1) time. The pulse train 19a of the analysis frame immediately before and the spectrum analyzer 2
From the spectrum coefficient 12 at time l input from
M samples of the influence signal 15 shown in FIG.
Output to. In the subtractor 5, from the weighting signal 14 to the influence signal
Corrected weighting signal 16 shown in FIG. 2 (f) with 15 subtracted
Is generated and output to the cross-correlator 7. In the cross correlator 7,
As shown in FIG. 2 (g), the cross-correlation function 17 of the correction weighting signal 16 and the impulse response function 13 is output to the (N + M) sample determination pulse extraction circuit 8. The autocorrelator outputs the autocorrelation function 18 of the impulse response function 13 to the M sample determination pulse extraction circuit 8 as shown in FIG. In the pulse extraction circuit 8, the operation of enlarging, reducing, and inverting the autocorrelation function 18 to a size equal to the maximum value of the crosscorrelation function 17, subtracting from the crosscorrelation function 17, and setting a pulse of equal size at that position is performed. By repeating the above, the pulse train 19 shown in FIG.
Output to.

〔The invention's effect〕

As described above, according to the present invention, a multi-pulse drive type speech coder including a spectrum analyzer, a weighting filter, a cross-correlator, an auto-correlator and a pulse extractor, and a weighted signal corrector By adding a subtraction unit and a subtractor, it is possible to provide a higher quality multi-pulse drive type speech encoder by removing abnormal noise generated at the boundary of the speech analysis frame by correcting the weighting signal.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. FIG. 2 is an operation waveform diagram of signals in the respective parts. 1 ... Input terminal, 2 ... Spectrum analyzer, 3 ... Weighting filter, 4 ... Weighting signal corrector, 5 ... Subtractor, 6 ... Autocorrelator, 7 ... Cross-correlator, 8 ...... Pulse extractor, 9 …… Pulse output terminal, 10 …… Spectrum coefficient output terminal, 11 …… Input voice signal, 12 …… Spectrum coefficient, 13 …… Impact response function, 14 …… Weighting signal,
15 …… Influence signal, 16 …… Correction weighting signal, 17 …… Cross-correlation function, 18 …… Autocorrelation function, 19 …… Transmitted pulse train, 19a …… Pulse train of the previous analysis frame.

Claims (1)

[Claims] 1. A spectrum analyzer (2) for obtaining a spectral coefficient and an impulse response function in an analysis frame of an input speech signal, and a weighting filter for weighting the input speech signal with the spectral coefficient to generate a weighted signal. (3) and a cross-correlator (7) for obtaining a cross-correlation function between the impulse response function and the weighted signal
And a speech encoder including an autocorrelator (6) for obtaining an autocorrelation function of the impulse response function, and a pulse extraction circuit (8) for extracting and outputting a pulse train from the crosscorrelation function and the autocorrelation function. In the above, a weighting signal correction circuit (4) for generating an influence signal from an analysis frame immediately before an analysis frame currently being analyzed from the pulse train and the spectral coefficient, and a correction weighting signal obtained by subtracting the influence signal from the weighted signal And a subtracter (5) for generating and inputting to the cross-correlator.