JPS60113300A - Voice 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 The present invention relates to a speech synthesis method, and particularly to a waveform synthesis method.

One of the speech synthesis methods is a waveform segment synthesis method.

This method focuses on the fact that adjacent audio waveforms have strong similarities. In the waveform segment synthesis method, when similar speech waveforms appear consecutively, one of the similar speech waveforms is selected as a representative segment, and this selected waveform is repeatedly used to synthesize the speech signal. The idea is to compress the amount of data. Methods for synthesizing waveforms of selected representative segments include PCM method, AD
Waveform encoding techniques such as the PCM (adaptive differential PCM) method can be used, but in particular, the deviation in synthesizing waveform segments using the ADPCM method is about 1/1 compared to when synthesizing waveform segments using the PCM method. Since waveform segments can be synthesized with a data amount of 2, it is a very effective means in terms of data amount compression.

The present invention relates to a waveform element synthesis method for synthesizing representative waveform elements using this ADPCM method, and repeating and using these representative waveform elements to synthesize speech.

The ADPCM method samples audio at the Nyquist frequency, and quantizes and equalizes the difference in amplitude values of audio waveforms between adjacent sampling points using an appropriate quantization width, and the quantization width is set at each sampling point. This is a method that adaptively changes the value according to the magnitude of the difference value. Therefore, when synthesizing representative waveform tree pieces using the ADPCM method, the quantization width is not constant within the waveform, and the quantization width at each sampling point depends on the immediately preceding ADPCM data. Further, in the ADPCM system, the sampling frequency is usually υf, which is 4 kHz to 8 kHz, depending on the relationship between sound quality and pit rate.

When synthesizing representative waveform segments using the AI) PCM method as described above, it is not possible to completely match the pitch period of the actual original waveform and the pitch period of the synthesized waveform. This is the pitch period 111p/of the waveform that can be synthesized.
If the sampling period is T1, then Tp'"" Ts
X N (N is the number of sampling points), T,
' can only take a value that is an integer multiple of T, and 1 maximum - L Ts
This is because it causes an error of . For example, when a waveform with a pitch period T as shown in FIG. 1 is sampled and synthesized at a sampling period T1, the end of the original waveform is at the sampling point 8t due to the influence of the pitch period error as described above.

and 821, the middle place is 1i. Therefore, set the last sampling point of the waveform to be synthesized to 5H1l+8
In either case, the amplitude of the final value of the synthesized waveform is set to 0.
It cannot be done. Furthermore, even if the pitch period of the original sound and the pitch period of the synthesized waveform perfectly match, the amplitude value at the last sampling point is the difference obtained by multiplying the amplitude value at the previous sampling point by the quantization width and the ADPCM code. However, depending on the value of the quantization width at the immediately preceding sampling point, it may not be possible to set the amplitude value at the last sampling point within one waveform to O.

Therefore, when synthesizing a representative waveform segment using the conventional ADPCM method as is, the amplitude value at the last sampling point of the representative waveform segment cannot be set to O. When repeated, the second
As shown in the figure, the amplitude center of the composite waveform changes.

This is because the ADPCM method is basically a differential encoding method that adds a difference value to the amplitude value of each sampling point to obtain the amplitude value of the next sampling point, so the final amplitude value of one waveform is not O. This is due to the fact that the error (ΔE in FIG. 1) is accumulated, and the amplitude center of the composite waveform changes.

If the amplitude center of the synthesized waveform changes, there is a possibility that the D/A converter that converts the waveform data into audio will overflow by the amount that the amplitude center has changed, and in that case, noise will be added to the audio, which is a serious drawback. occurs.

Tree brother! The purpose of j14 is A. Sound 7+f synthesis, in which voice can be synthesized without changing the amplitude center of the synthesized waveform even if the representative waveform elements synthesized using the D-P CM method are used repeatedly to synthesize voice. The purpose is to provide a method.

An embodiment of the present invention will be described below with reference to FIG. 3 of the drawings. In this synthesis method, for each sampling point, the amplitude value is divided by the difference value from the previous sampling point, and the difference value is calculated by 1M.
Rather than obtaining the i value, we use a method to force the 4h' at the sampling point 10 times before to 0 (itf is a means of forcing an innocent person to 0. Therefore, the waveform element synthesized by the C1 method is Even if the waveform shown in Figure 3 is synthesized using the conventional method, the amplitude value at the last sampling point will be at the position indicated by 815, but the amplitude center of the synthesized waveform will not change at all even if the waveform shown in Figure 3 is synthesized using the conventional method. If the method is used, the amplitude value at the last sampling point in one waveform will be 0 regardless of S14.S and l+, so the final sampling point in one waveform will be 5lfi'.In this case, the pitch period of the original waveform An error of ΔT occurs between the pitch period of the synthesized waveform, but this can be reduced by setting the sampling period T@ to an appropriate value. As a circuit, this can be realized by a circuit (for example, a counter) that detects that it is the final sampling point and a circuit that sets '0' as data at that point.

As explained above, according to the present invention, even if a representative waveform segment is created using the ADPCM method and voice is synthesized by repeating the representative waveform segment, the amplitude center of the synthesized waveform does not change. You can easily create audio waveforms like this. Furthermore, according to the present invention, there is no need to have data for outputting the last sampling point of the representative waveform segment used for repeating, so the bit rate during synthesis can be improved accordingly. .

[Brief explanation of drawings]

Figure 1 is a waveform diagram 1 to explain the conventional speech synthesis method.
Figure 2 is a waveform diagram showing a waveform synthesized by the conventional a-voice synthesis method, and Figure 3 is a waveform diagram for explaining the voice synthesis method of the present invention. \-1-

Claims (1)

[Claims] A speech synthesis method in which representative waveform segments are synthesized using an ADPCM method, characterized in that the amplitude value at the last sampling point of the synthesized waveform is forcibly set to 0.