JPS62245300A - Soundless section generating 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] Pincer Natsume 1! The present invention relates to a silent interval generation method in speech synthesis using an LSP speech synthesis filter.

In the speech analysis and synthesis method that uses the storm spectrum pair (LSP) as a parameter, based on the generation model, the LS
Synthesized speech is obtained by applying the P coefficient and the sound source waveform to a speech synthesis filter. As the sound source waveform, generally a pulse train (voiced sound part) matching the pitch period and white noise (unvoiced sound part) are used.

Conventionally, in order to generate a silent section such as the front part of a silent plosive, methods such as simply setting the sound source waveform to O or initializing the synthesizer by resetting it have been used. However, just by setting the sound source to O, the synthesized speech waveform does not immediately become O, but due to the variables stored inside the LSP speech synthesis filter.

Some kind of waveform may continue to be output and may be heard as an abnormal sound, and if one synthesizer is reset, the synthesized speech waveform may be abruptly terminated, resulting in a popping sound.

Purpose The present invention was made in view of the above-mentioned circumstances.
In particular, in the LSP speech synthesis method, good silence intervals (
This was done for the purpose of generating plosive sounds (such as the front part of a plosive sound).

In order to achieve the above object, the present invention includes an anti-resonant circuit having a plurality of zero points on a unit circle on a complex plane in a feedback loop, and controls the anti-resonant frequency of this anti-resonant circuit. In an LSP voice synthesis filter that synthesizes various voices by changing the spectral envelope characteristics of voice, the LSP voice synthesis filter has means for gradually attenuating the amplitude of the synthesized voice,
Let the sound source waveform in the silent section be O, and at the same time, LSP
This allows parameters to be spaced at equal intervals. An explanation will be given below based on one embodiment of the present invention.

FIG. 1 is an electrical block diagram for explaining one embodiment of the present invention. In the figure, 1 is an adder, 2 is a delay device, and 3 is a multiplier. The internal configuration of the speech synthesis filter is shown, and FIG. 2 shows parameter changes during a silent section. ωi (1≦i≦4) is an LSP coefficient corresponding to the frequency from 0 to π, C1 (1≦i≦4) is a parameter given to the synthesis filter, and Ci = −2cosω
There is a relationship that makes it i. Making ωi evenly spaced is equivalent to setting the transfer characteristic of the LSP speech synthesis filter to 1. In other words, when a sound source waveform is given to the LSP speech synthesis filter by resetting parameters in which ωi is made equally spaced. , the sound source waveform will be output as is. Therefore, if the sound source is set to 0 in this state, a good silent section can be generated.

FIG. 3 is a diagram for explaining another embodiment of the present invention, and this embodiment variably controls the magnification of a multiplier with a magnification of -172 included in the feedback loop of the LSP speech synthesis filter. This is what I did.

Figure 4 is a diagram showing a method of controlling the magnification during a silent section. During the silent section, the input to the feedback loop of the filter is gradually attenuated to reduce its function as a filter, so that a good silent section can be generated. I have to.

FIG. 5 is a diagram for explaining another embodiment of the present invention,
(A) is an overall circuit diagram, and (B) is a detailed diagram of the four parts in (A). In this embodiment, each delay element 2 included in the LSP speech synthesis filter has a constant magnification K (K 1) can be added using switch 5. During normal audio output, the delay elements operate without attenuation, and during silent periods, switch 5 allows each delay element to have an attenuation term.
Switch and use. By this operation, the value of the internal variable in the feedback loop of the filter gradually approaches O, so that a good silent section can be generated. Note that the speed at which the output audio attenuates at the end of the silent section can be adjusted by the value of K.

FIG. 6 is a diagram for explaining another embodiment of the present invention,
In this embodiment, a multiplier whose magnification can be controlled is provided at the output end of an LSP speech synthesis filter, and a method for controlling the magnification of this multiplier is shown in FIG.

Since this embodiment controls the amplitude of the filter output, it is possible to generate a silent section without having to set the input sound source to O as in the other embodiments.

As is clear from the above explanation, according to the present invention, the LS
It becomes possible to attenuate the output of the P speech synthesis filter at an appropriate speed, and it is possible to generate a good silent section.

[Brief explanation of drawings]

FIG. 1 is an electrical block diagram for explaining an embodiment of the present invention, FIG. 2 is a diagram showing parameter changes in a silent section, and FIG. 3 is an electrical block diagram for explaining an embodiment of the present invention. FIG. 4 is an electrical block diagram for explaining the method of controlling the magnification in a silent section, and FIGS. 5 and 6 are electrical block diagrams for explaining other embodiments of the present invention. 7 are diagrams for explaining a method of controlling the magnification of a single multiplier. 1... Adder, 2... Delay device, 3... Multiplier, 4
...Combination circuit of delay unit and multiplier, 5...Switch.

Claims (5)

[Claims] (1) An anti-resonant circuit having multiple zero points on a unit circle on a complex plane is included in the feedback loop, and by controlling the anti-resonant frequency of this anti-resonant circuit, the spectral envelope characteristics of the voice are changed, and various LSP that synthesizes the audio of
1. A silent interval generation method, in a speech synthesis filter, comprising means for gradually attenuating the amplitude of synthesized speech. (2) In the silent part where the input sound source is 0, the LS
The silent interval generation method according to claim 1, wherein the transfer characteristic of the P speech synthesis filter is set to 1, so that the input sound source is output as is. (3) Claim (1) characterized in that a -1/2 multiplier included in the feedback loop of the LSP speech synthesis filter is made variable, and the multiplier gradually approaches 0 in silent parts. Silent section generation method described. (4) A constant attenuation rate can be added to each delay element included in the feedback loop of the LSP speech synthesis filter by a switch, and the switch is turned on in a silent part to cause the delay element to have attenuation. A silent section generation method according to claim (1). (5) A silent section according to claim (1), characterized in that an attenuator whose magnification can be controlled is provided at the output section of the LSP speech synthesis filter, and the magnification is gradually brought closer to 0 in the silent section. Generation method.