JP3200872B2 - Code-driven linear predictive speech coding - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code-driven linear predictive speech coding method, and more particularly to a sound quality improving method using pitch information.

[0002]

2. Description of the Related Art In a conventional code-driven linear predictive speech coding system using pitch information, the pitch is determined by the sampling interval of the input speech, so that the pitch accuracy is the sampling interval of the input speech.

[0003]

In the above-described conventional code-driven linear predictive speech coding using pitch information, the precision of the pitch is the sampling interval of the input speech, so that the precision of the pitch is not sufficient. Not only that, the accuracy of the codebook search is not sufficient, and there is a limit that the improvement of sound quality is limited.

[0004] It is an object of the present invention to provide a code-driven linear predictive speech coding system with improved sound quality.

[0005]

In order to achieve the above object, a code-driven linear predictive speech coding system according to the present invention comprises:
A code-driven linear prediction speech coding method using pitch information, wherein an input speech is interpolated to create a speech signal with an increased number of samples, and a linear prediction filter coefficient is calculated from the speech signal with the increased number of samples. and calculating a pitch information, said constitutes a pitch reproduction filter from the pitch information, and constitutes a linear prediction filter from the linear prediction filter coefficients, the codebook same as if no interpolation
The position where the code is set is the input before interpolation
Only the audio sample position is used, and the pitch period is
Sampling interval, and the pitch reproduction filter is
The drive signal driven by the mode is placed at the interpolated sampling position.
By also make synthetic speech to the cascade filter is driven by the code of the line <br/> Predictor filter and the pitch reproduction filter searches the codebook to be closest to the original speech, a predetermined number of obtains the code information, the transmitting encodes the linear prediction filter coefficients and the pitch information the code information, decodes the received code at the receiving end, the <br/> codes and the linear prediction filter coefficients and the pitch information restore the information, the first drive signal is synthesized by reading the code from the code <br/> codebook according to the code information, constitute the pitch reproduction filter from the pitch information, in the first drive signal by driving the pitch reproduction filter, the second driving signal by combining, forming the linear prediction filter from the linear prediction filter coefficients And it reproduces the audio by thinning out a signal obtained by driving the linear prediction filter in the second drive signal.

[0006]

According to the present invention, an audio signal having a fine sampling interval is created by appropriately interpolating an input voice, and a sample of the input voice before interpolation is generated by using fine pitch information of the sampling interval calculated from the audio signal. By performing a codebook search while setting a code only at a position, the sound quality is improved with almost no change in the amount of transmission information as compared with the case without interpolation.

[0007]

Next, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a code-driven linear predictive speech coding system according to the present invention.

The code-driven linear predictive speech coding method of the present embodiment will be described with reference to FIG. The audio signal X (n) (n = 0, 1, 2,..., N−1) applied to the input terminal 10 is supplied to the interpolator 1 where it is interpolated. Interpolated audio signal X _I (n) (n = 0, (1/2), 1, 1 (1 /
2),..., N−1) are supplied to the linear prediction analyzer 2, the pitch extractor 3, and the codebook searcher 5.

In the linear prediction analyzer 2, the interpolated speech signal X _I (n) is subjected to linear prediction analysis to obtain a coefficient α of a linear prediction filter, which is supplied to a codebook searcher 5 and an encoder 6.

The pitch extractor 3 extracts pitch information from the interpolated audio signal, and extracts the pitch information from the interpolated speech signal.
Is supplied to the encoder 6.

The codebook searcher 5 forms a pitch reproduction filter from the pitch information, forms a linear prediction filter from the linear prediction filter coefficient α, and generates a synthesized speech obtained by driving a cascade filter of the pitch reproduction filter and the linear prediction filter by a code. Is searched for from the code book 4 so that a is closest to the original voice, a predetermined number of pieces of code information are obtained,
It is supplied to the encoder 6.

In the encoder 6, a linear prediction filter coefficient α
, Pitch information and code information are encoded and output from the encoder output terminal 11 to the transmission path.

On the receiving side, the encoded linear prediction filter coefficient α, pitch information and code information input to the decoder input terminal 30 are decoded by the decoder 21, and the linear prediction filter coefficient α is supplied to the linear prediction filter 25. The pitch information is supplied to a pitch reproduction filter 24,
The code information is supplied to the drive signal synthesizer 23.

The drive signal synthesizer 23 synthesizes a first drive signal by reading a code from the code book 22 in accordance with the code information, and supplies the first drive signal to the pitch reproduction filter 24.

In the pitch reproduction filter 24, the second drive signal is used to drive the pitch reproduction filter so that the second
Are combined and supplied to the linear prediction filter 25.

The linear predictive filter 25 drives the linear predictive filter with the second drive signal to generate a synthesized speech at the interpolated sampling interval, and
To supply.

The decimation unit 26 decimates the synthesized speech at the interpolated sampling interval, creates a synthesized speech at the sampling interval before interpolation, and outputs it from the decoder output terminal 31.

Next, each part of the multi-pulse encoding system of this embodiment will be described with reference to FIG.

(1) Pitch extraction The pitch is extracted by input speech X (n) (n = 0, 1,...,
N-1) (shown in FIG. 2 (a)) interpolated audio signal X _I
(N) (n = 0, (1/2), 1, 1 (1/2),...,
N-1) (shown in FIG. 2 (b)), the sampling interval becomes small, and if interpolation is performed properly,
More accurate pitch is required.

(2) Codebook Search A codebook is searched so that a synthesized speech obtained by driving a cascade filter of a pitch reproduction filter and a linear prediction filter by a code becomes closest to the original speech. Here, the same codebook as in the case where no interpolation is performed is used, and the code is formed only at the sample position of the input voice X (n) before interpolation (shown in FIG. 2C). However, since the pitch period is the interpolated sampling interval, the drive signal D (n) (shown in FIG. 2D) in which the pitch reproduction filter is driven by the code also stands at the interpolated sampling position. In this way, the synthesized speech in which the cascade filters of the pitch reproduction filter and the linear prediction filter are driven by the code can be made closer to the original sound as compared with a case where a codebook is searched without interpolation.

The same code book as that used when no interpolation is performed is used, and the code is set only at the sample position of the input voice X (n) before the interpolation, so that the code information to be transmitted does not change. As the transmission information, only the pitch information amount increases, but the pitch information is sufficiently smaller than the code information, and the transmission information amount hardly changes.

[0023]

As described above, according to the present invention, an audio signal having a small sampling interval is generated by appropriately interpolating an input audio signal, and interpolation is performed using fine pitch information of the sampling interval calculated from the audio signal. By performing a codebook search while setting a code only at the sample position of the input voice before the input, the sound quality can be improved with almost no change in the amount of transmitted information as compared with a case without interpolation.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a code-driven linear predictive speech coding method according to the present invention.

FIG. 2 is a diagram showing signals of respective units in the code-driven linear predictive speech coding method according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Interpolator 2 Linear prediction analyzer 3 Pitch extractor 4 Codebook 5 Codebook searcher 6 Encoder 10 Encoder input terminal 11 Encoder output terminal 21 Decoder 22 Codebook 23 Drive signal synthesizer 24 Pitch reproduction filter 25 Linear Prediction filter 26 decimator 30 decoder input terminal 31 decoder output terminal

Claims (1)

(57) [Claims] 1. A code-driven linear predictive speech coding method using pitch information, wherein a speech signal having an increased number of samples is created by interpolating an input speech, and the speech signal having the increased number of samples is generated. Calculate the linear prediction filter coefficient and pitch information from, configure a pitch reproduction filter from the pitch information,
A linear prediction filter is constructed from the linear prediction filter coefficients, and the same codebook as that used when no interpolation is performed is used.
The position of the input sound is the sample position of the input sound before interpolation.
The pitch period is the same as the interpolated sampling interval.
And a drive in which the pitch reproduction filter is driven by the code.
The motion signal is set at the interpolated sampling position.
Ri, the pitch reproduction filter and synthesized speech to the cascade filter is driven by the code of the linear prediction filter searches the codebook to be closest to the original speech, obtains the code information of a predetermined number of said linear prediction wherein the filter coefficients and the pitch information is encoded code information and transmitting, and decoding the received symbols at the receiving side to restore the code information and the linear prediction filter coefficients and the pitch information, the codebook according to said code information the first drive signal is synthesized by reading the code from, constitute the pitch reproduction filter from the pitch information, the first
By driving the pitch reproduction filter in the drive signal, the second driving signal by combining the constituting said linear prediction filter from the linear prediction filter coefficients, driving the linear prediction filter in the second driving signal A code-driven linear predictive speech coding method for reproducing speech by thinning out a signal obtained by performing the above.