JPS6135499A - Voice analysis/synthesization system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a speech analysis and synthesis method, and in particular to a speech analysis and synthesis method that mainly uses line spectrum pairs (Line Sp) by spectral coding.
ectrum Psir + below, LSP,! :say)
method, partial autocorrelation coefficient (pBrtiBl-Buto-C
The present invention relates to a speech analysis and synthesis method using PARCOR (hereinafter referred to as PARCOR) method.

[Prior art]

Conventionally, in the speech analysis and synthesis methods such as PARCOR and LSP, the original speech is compressed into several parameters to represent it, so that one synthesized speech may contain abnormal sounds that are not in the original speech.

[Purpose of the invention]

An object of the present invention is to provide a speech analysis and synthesis method that can prevent or reduce the occurrence of abnormal sounds.

[Structure of the invention]

In the present invention, when an abnormal electric sound occurs in the above speech analysis and synthesis method, the L8P coefficient wi, the PARCOR coefficient ki
Focusing on the fact that the acoustic parameters such as
By smoothing the OR coefficient ki, the occurrence of abnormal noise is prevented or reduced.

That is, the speech analysis and synthesis method of the present invention includes an acoustic parameter analysis processing means that forms acoustic parameters for each frame period from the original speech or other convertible characteristic parameters that have already been obtained; It is determined whether the absolute value of the difference between the value of the acoustic parameter and the acoustic parameters of the previous and subsequent frames is larger than a predetermined value, and when at least one of the values is larger, the acoustic parameter is The weighting of the acoustic parameters is characterized by including a smoothing processing means for smoothing by averaging.

〔Example〕

Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a speech analysis and synthesis method in which the present invention is applied to an LSP method.
0 and a speech synthesizer 12.

Next, the functions of each block will be explained.

The original audio is converted into a digital signal by an A-filter converter 1. The window processing device 2 performs data window processing and calculation of linear prediction coefficients. The data processing device 3 is the window processing device 2
The data processing unit receives the digital signal subjected to data window processing and performs data processing such as pitch extraction, voiced/unvoiced, and voiced/silent determination. On the other hand, the linear prediction coefficients obtained by the window processing device 2 are converted into LSP coefficients by the conversion processing device 4.

The pitch, voiced/unvoiced, and voiced/silent signals formed by the data processing device 3 and the LSP coefficients calculated by the conversion processing device 4 are written into the mesh 5. At the same time, the processing contents of the previous frame are written from the memory 5 to the memory 6, and the processing contents of the frame before the previous one are written from the memory 6 to the memory 7. The subtraction processing device 8 subtracts the difference between the LSP coefficients of the memories 5 and 6, and the device 9 subtracts the difference between the LSP coefficients of the memories 6 and 7.
The difference between the SP coefficients is calculated and output to the smoothing processing device 10. In the smoothing processing device 10 (provided in the speech analysis device or speech synthesis device), weighting is performed using the LSP coefficients of the previous and subsequent frames stored in the memories 5 and 7 only when the coefficients are larger than the subtraction processing device. By taking L
Find the SP coefficient.

That is, the smoothing processing unit [10] weights i-1 by coefficients α1°α2, . Smoothing is performed using α3 according to the following equation (1).

Less/#: Margin 1/White , -1 Indicates that it is a system.

In this case, Wn+1 is the LSP stored in memory 7.
The coefficient, Wn is the LSP coefficient stored in memory 6, -+1 is the LSP coefficient stored in memory 5, Wn
' is obtained by smoothing the LSP coefficient wB stored in the memory 6.

The thus smoothed LSP coefficient wn' and the pitch, voiced/unvoiced, and voiced/silent signals stored in the memory 6 are written in the memo 1711. If smoothing is not necessary, the contents of memory 6 are written to memory 11 as they are.

The speech synthesizer 12 uses the pitch output from the memory 11,
Synthesized speech is formed using voiced/unvoiced, voiced/unvoiced signals, and LSP coefficients. This synthesized voice is pronounced through a speaker (not shown).

Since sudden changes in the LSP coefficients are suppressed in this way, the occurrence of abnormal sounds is prevented or significantly reduced.

This embodiment is for the L8P method, but the present invention is based on the PAR
It is also applicable to the COR method and other acoustic parameter methods.

〔Effect of the invention〕

As explained above, the present invention utilizes LSF coefficients, PARCO
Since acoustic parameters such as R coefficients are smoothed only when they change rapidly between frames, sudden changes in acoustic parameters between frames are suppressed, preventing or significantly reducing the occurrence of abnormal sounds. be able to.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the speech analysis and synthesis method using the LSP method of the present invention. 1: A/l) converter, 2: window processing device, 3: data processing device, 4! Conversion processing device, 5.6, 7, 11: Memory, 8.9: Subtraction processing device, 1o! Smoothing processing device, 12-Speech metallization device.

Claims (1)

[Scope of Claims] 1. Acoustic parameter analysis processing means for forming acoustic parameters for each frame period from the original speech or other transformable feature parameters that have already been obtained; Determine whether the absolute value of the difference between the value of the parameter and the acoustic parameters of the previous and subsequent frames is larger than a predetermined value, and when at least one of the values is larger, the acoustic parameter is used as the acoustic parameter of that frame and the previous and subsequent frames. A speech analysis and synthesis method characterized by having a smoothing processing means for smoothing by a weighted average of parameters. 2. The speech analysis and synthesis method according to claim 1, wherein the smoothing processing means is provided in the speech analysis device. 3. The speech analysis and synthesis method according to claim 1, wherein the smoothing processing means is provided in the speech synthesis device.