JPS603000A - 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 speech synthesis method using a spectral coding method.

Natural voice? Sampling and constant analysis period (frame period)
Various types of information necessary for speech synthesis (hereinafter referred to as speech parameters) are extracted from sample values of speech within a single mouth or natural speech, and are artificially created similar to the extracted speech parameters. In speech synthesis using a spectral coding method using parameters, a method is known in which the speech parameters used for synthesis are updated at the same time period as the frame period during one synthesis. However, this -periodic voice parameter update does not take into account all the transient characteristics of the synthesis filter used for speech synthesis, and the synthesis filter becomes unstable immediately after the voice parameter update, resulting in unnatural speech synthesis. become. The naturalness of the sweet-sour voice is particularly bad when the voice parameters change rapidly in the M7” dorsal range.

An object of the present invention is to provide a speech synthesis method that generates all synthesized sounds with excellent naturalness.

The disadvantage of the present invention is that the excitation signal is input to the t-filter and all the filter coefficients of this filter are changed to synthesize the desired voice. The system is characterized in that, at least when the excitation signal having a predetermined characteristic is input to the filter, the filter coefficients are updated in synchronization with the input of the excitation signal.

According to the present invention, depending on the characteristics of the excitation signal of the filter (for example, whether the filter is driven by a 1M voice source or a silent sound source), the filter is applied to the synthesis filter. By changing all the coefficients of the synthesis filter at the input timing of the excitation signal, for example, in synchronization with the pitch cycle when driving a voiced sound source, the filter blade just before the excitation number 1M of the filter (at the end of the pitch cycle) is the most attenuated. Therefore, it is possible to update all filter coefficients at the most appropriate time for the change. When driving a silent sound source, the sound obtained as a filter blade takes only a small value like natural speech, so it is changed when a predetermined length or a time length included in the audio parameter has elapsed. Conventional methods may also be used.

According to the present invention, the filter coefficients of the synthesis filter are changed in accordance with the characteristics of the excitation signal of the t1 filter and in synchronization with the input of the excitation signal of the filter to the synthesis filter. Even when using parameters, are they artificially created? Even when using it, does the synthesis filter become unstable immediately after updating the audio parameters? Therefore, it is possible to provide a speech synthesis method that can generate synthesized sounds with excellent naturalness and high quality.

Is there a drawing below? The present invention will be explained in more detail with reference to the drawings.

Is Fig. 1 an embodiment of the present invention? FIG. The speech synthesis device 10 includes a speech parameter storage circuit 20 (such as ROM or RAM) that stores speech parameters used for synthesis, and a filter coefficient input circuit 30 that inputs filter coefficients output from the storage circuit 20. A voiced sound source circuit 40 that generates all voiced excitation signals, an unvoiced sound source circuit 50 that generates unvoiced excitation polarization, and a filter excitation signal manual circuit 60 that inputs the generated filter excitation polarization.
, a synthesis filter 70 that performs voice synthesis in response to the filter coefficients and excitation signals input from the filter coefficient human power circuit 30 and the filter excitation signal human power circuit 60, and a voice output terminal 80 that outputs a synthesized sound 1M,
Reading audio parameters from the storage circuit 20 and filter coefficient input circuit 30. Voiced sound source circuit 40. A control circuit 90 that controls all operations of the silent Shingen circuit 50° filter excitation signal power circuit 60 and the synthesis filter 70, and a command input terminal 100 that manually issues instructions to the control circuit 90 to start and stop voice synthesis operations We are prepared.

The speech synthesis device 10 is activated by the control circuit 90 in accordance with a command inputted from the command input terminal 100, and the entire synthesis process is started. Control circuit 901''l: Specifies all the addresses of the voice parameters. The memory circuit 20 outputs the filter coefficient t of the synthesis filter 70, which is one of the voice parameters specified, to the filter coefficient human power circuit 30, and performs M voice/voice discrimination. , pitch period, sound source amplitude, etc., to the control circuit 90.
Send to. The control circuit wr90 operates the voice sound source circuit 40 according to the received 7'CV voiceless discrimination data. Alternatively, the pitch period and sound source amplitude data are sent to the silent sound source circuit 50. When the voice sound source circuit 40 is selected, the control circuit 90 closes the filter coefficient circuit 30, manually inputs the filter coefficient to the synthesis filter 70, and inputs the filter coefficient to the voiced sound source circuit 40.
The filter excitation polarization according to the source amplitude?
At the same time as generating the filter excitation signal, the human power circuit 60 is closed and the excitation signal? A large amount of power is applied to the synthesis filter 70.

The synthesis filter 70 performs synthesis according to the operation signal (clock note) outputted from the control circuit 90 and outputs the synthesized sound to the audio output terminal 80 . On the other hand, the control circuit 9
0 repeats the transmission of operation No. 11 that outputs a large amount of power to the synthesis filter 70 at regular time intervals, counts the number of times of transmission, and determines whether it is an integral multiple of the pitch period. To detect. '! ,′
fc, the control circuit 90 counts the total synthesis time from the number of transmissions and a fixed transmission time interval, and when the synthesis time matches the frame period, counts the number of transmissions exceeding the frame period? Start. When the counting result of the number of sending out matches an integral multiple of the pitch period, the control circuit 90 detects whether the counting result of the number of excess sending out is O, and if it is 0, the control circuit 90 starts the voiced sound source circuit 4 again.
0, the filter excitation signal input circuit 6
0? ! - close, powering the excitation signal onto the synthesis filter 70; If the counting result of the number of excess transmissions is other than 0, specify a new audio parameter address to the control circuit 90,
Result of counting the number of excess transmissions? After setting the result as the number of transmissions, is the number of excess transmissions? Return to 0.

On the other hand, the storage circuit 20 stores audio parameters of the address newly designated by the control circuit 90.

Filter coefficient of synthetic filter? It outputs data to the filter coefficient input circuit W&30 and sends data such as 1MP unvoiced discrimination, pinch period, sound source amplitude, etc. to the control circuit 90, and performs voice synthesis without changing any of the above operations.

As described above, in this implementation, instead of changing the filter coefficients for each frame period, all the filter coefficients can be changed at the end of the pitch period that first passes after the frame period, and the voice synthesis that returns damped vibrations v<b for each pitch period is possible. Changing filter coefficients at the most appropriate point in the filter? By doing this, it is possible to synthesize high-quality speech with excellent naturalness, in contrast to the conventional method in which all filter coefficients are changed every frame period. Moreover, it is clear from the drawings that the two-wire system can be realized with a relatively small amount of circuitry and can be easily manufactured using well-known semiconductor technology.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one implementation of the present invention. 10...Speech synthesis device, 30...Filter coefficient human power circuit, 40...Stone sound source circuit,
50... Silent episode! , 60... Filter excitation signal human power circuit, 80... Back door output terminal, 100... Command input terminal.

Claims (1)

[Claims] In a method in which the excitation signal is input to an excitation iq'e filter and the filter coefficients of this filter are all updated to synthesize the required speech, at least when the excitation signal with a predetermined characteristic is input to the filter, the filter coefficients are The update shall be performed in synchronization with the input of the excitation signal.
Characteristic voice synthesis method.