JP2580123B2 - Speech synthesizer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a speech synthesis device, and more particularly, to a speech synthesis device using residual driving.

2. Description of the Related Art Conventionally, in a speech synthesis method of obtaining a synthesized speech by inputting a spectrum envelope parameter (LPC, PARCOR, LSP, etc.) and a sound source signal to a synthesis filter, as a driving sound source, an impulse train in a voiced portion, a cut out one pitch Periodic residual waveform,
For non-speech parts such as average or representative residual waveforms,
White noise (such as M-sequence) has been used. Although the above-described prior art has an effect of improving sound quality by using the residual signal, however, since the use of the residual signal is performed partially or selectively, sufficient clarity and sound quality are obtained. There was a drawback that sometimes it could not be obtained.

Object The present invention has been made in view of the above circumstances, and in particular, by generating a driving sound source signal from a residual signal obtained by inverse filtering, it is possible to generate high-quality synthesized speech as compared with the related art. It was made for the purpose of obtaining.

In order to achieve the above object, the present invention provides a speech synthesizer for synthesizing a speech waveform using a spectrum envelope parameter extracted from a speech waveform and a residual waveform extracted by inverse filtering processing. A spectrum envelope parameter storage unit for storing the obtained spectrum envelope parameter, a residual waveform storage unit for storing a residual waveform obtained from the parameter and a residual waveform obtained by deforming the residual waveform, and a residual waveform storage unit obtained by inverse filtering. Directly using the obtained residual waveform as a direct drive sound source, or a drive sound source generation control unit that generates control data to specify whether to use the transformed sound source, according to control data from the drive sound source generation control unit, A driving sound source generating unit that generates driving sound source data from the residual waveform data read from the residual waveform storage unit; And a synthesis filter for synthesizing the driving sound source data created by the generation unit and the parameters from the spectrum envelope parameter storage unit. In the bursty unvoiced sound part, the residual waveform obtained by inverse filtering is directly used as the driving sound source. In other unvoiced sound portions, a waveform obtained by deforming the extracted residual waveform is used as a driving sound source. Hereinafter, a description will be given based on examples of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of the present invention, in which 1 is a spectrum envelope parameter storage unit, 2 is a residual waveform storage unit, 3 is a driving sound source generation control unit, and 4 is A driving sound source generating unit, 5 is a synthesis filter unit, 6 is a D / A converting unit, and 7 is an output terminal.In the present invention, a residual waveform generated when a spectral envelope parameter is extracted as a driving sound source waveform is directly or modified. Used after processing.

One pitch of the residual waveform obtained by inverse-filtering the original residual waveform using the spectral envelope parameters (LPC, LSP, etc.) as the driving sound source waveform and transforming only the phase characteristic of the predicted residual waveform to the same phase What is necessary is just to connect repeatedly by using the period, to perform shaping such as determining the amplitude, and use it. In the non-voice portion, the predicted residual waveform may be directly used. In the silent part, control is performed so that the residual waveform is directly used as the driving sound source for the rupture part, and multiple single-time representative residuals extracted from the residual waveform are randomly connected to the friction part as the driving sound source. May be.

In FIG. 1, a spectrum envelope parameter extracted from an audio waveform, a residual waveform obtained by inverse filtering the original audio waveform using the parameter, a residual waveform obtained by deforming the residual waveform, and the like are each a spectrum. It is stored in the envelope parameter storage unit 1 and the residual waveform storage unit 2.
The driving sound source generation unit 4 converts the residual waveform data read from the residual waveform storage unit 2 into control data (pitch, amplitude, prediction residual waveform or a modified residual waveform, etc.) from the control unit 3. ) Is selected, connected, and shaped to generate driving sound source data, which is sent together with the spectral envelope parameter data to the synthesis filter unit 5 which is a vocal tract model circuit, and the output of the synthesis filter unit 5 is passed through the D / A conversion unit 6. The signal is converted into an analog signal, and a synthesized voice signal is output to the output terminal unit 7.

Advantages As is apparent from the above description, according to the present invention, the sound quality and intelligibility of synthesized speech are improved by using the residual waveform generated when extracting the spectral envelope parameter from the natural speech waveform, directly or by using a modified process. Can be done.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining an example of a system used for implementing a speech synthesizer according to the present invention. 1 ... Spectral envelope parameter storage unit, 2 ... Residual waveform storage unit, 3 ... Drive sound source generation control unit, 4 ... Drive sound source generation unit, 5 ... Synthesis filter, 6 ... D / A conversion unit, 7 Output terminal.

Claims (1)

(57) [Claims] 1. A speech synthesizer for synthesizing a speech waveform using a spectrum envelope parameter extracted from a speech waveform and a residual waveform extracted by inverse filtering, stores a spectrum envelope parameter extracted from the speech waveform. A spectrum envelope parameter storage unit, a residual waveform storage unit for storing a residual waveform obtained from the parameter and a residual waveform obtained by subjecting the residual waveform to a deformation process, and directly driving a residual waveform obtained by inverse filtering A driving sound source generation control unit for generating control data for specifying whether to use directly as a sound source or to use the sound source after transformation processing, and read out from the residual waveform storage unit according to control data from the driving sound source generation control unit. A driving sound source generating unit that generates driving sound source data from the generated residual waveform data, and a driving sound generated by the driving sound source generating unit. Data, and a synthesis filter for synthesizing the parameters from the spectrum envelope parameter storage unit.In the bursty unvoiced part, the residual waveform obtained by the inverse filtering is directly used as a driving sound source, and in other unvoiced parts, A speech synthesizer characterized by using a waveform obtained by transforming an extracted residual waveform as a driving sound source.