JPH02230300A - Voice synthesizer - Google Patents

Abstract

PURPOSE:To prevent energy from being converted excessively by providing the voice synthesizer with a pitch varying device and determining a pulse excitation point at or nearly at a pitch pulse period. CONSTITUTION:When a voice synthesizing filter 13 is driven by a modeled sound source with an impulse string based upon the pitch period or a white noise to generate an input voice signal, the pitch pulse varying device 18 which varies the generation period of a pitch pulse train based upon the pitch period is provided and a pitch pulse position which is a pulse excitation point varies at a period determined by the pitch pulse varying device 18. Consequently, the convergence of energy is evaded and a synthetic voice which is close to natural one can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a speech synthesizer that synthesizes spectral envelope information and sound source information analyzed by a speech analyzer.

[Conventional technology]

Conventionally, this type of speech synthesizer inputs spectral envelope information indicating the macroscopic structure of an input speech signal and sound source information indicating the fine structure, and reproduces a human-powered speech signal based on these analysis information. That is, such a speech synthesizer is equipped with an all-pole digital filter as a speech synthesis filter that drives spectral envelope information transmitted as analysis information with sound source information.

Spectral envelope information is usually obtained by converting the input audio signal to LPG.
analysis (linear predictive analysis; Linear Predicti
The filter coefficients are linear prediction coefficients such as the α parameter and the K parameter determined by

On the other hand, the sound source information is an impulse sequence based on the pitch period (
(pitch pulse train), voiced/unvoiced/silent information, other audio power, etc., and the waveform information contained in the sound source information is cut out, and the pitch period of the sound source, voiced/unvoiced/silent information, audio power, etc. are modeled, and the voice synthesis I am trying to drive the filter. That is, a voiced sound source is driven by an impulse train of its pitch period, and a voiceless sound source is represented by a modeling representation using white noise to drive the speech synthesis filter.

[Problem to be solved by the invention]

However, in a conventional speech synthesizer that drives a speech synthesis filter using analysis information without waveform transmission, compared to a waveform transmission type speech synthesizer such as a multipulse vocoder, it is essentially Phase information is lacking, and pitch period information is also prone to ambiguity. In particular, since the speech synthesis filter is driven using the pitch pulse position generated in response to the pitch period information transmitted from the analysis side to the synthesis side as a pulse excitation point, energy is excessively concentrated at the pulse excitation point. Excitation using the pitch period causes concentration of energy at regular intervals, resulting in a synthetic sound quality that is mechanical and lacks naturalness.

[Means to solve the problem]

The present invention provides a method for controlling the generation period of a pitch pulse train (impulse train) based on the pitch period when synthesizing an input speech signal by driving a speech synthesis filter with a modeled sound source using an impulse train based on the pitch period or white noise. It is characterized by being provided with a variable pitch pulse device.

[Effect]

Since the pitch pulse variable device changes the pitch pulse position, which is the pulse excitation point, according to the period determined by the pitch pulse variable device, concentration of energy can be avoided and a synthesized sound close to naturalness can be obtained.

〔Example〕

Hereinafter, the present invention will be described in detail with reference to an embodiment shown in the drawings.

FIG. 1 is a block diagram showing an embodiment of the speech synthesizer of the present invention. The demultiplexer 11 manually inputs analysis information of the audio signal to be synthesized via the transmission line 12. This analysis information is a multiplexed signal of spectral envelope information and sound source information by a voice analyzer, and includes LPG coefficient data a as spectral envelope information, voiced/unvoiced silence information b as sound source information, pitch period information C, and short-time voice power. Data d is included.

Here, the voice analyzer that analyzes the voice signal is composed of an LPG analyzer, a pitch extractor, a voiced/unvoiced/silence discriminator, a power meter, etc., and stores the analyzed analysis information in a memory circuit and uses a multiplexer etc. as appropriate. Combine and multiplex
This is transmitted encoded, sent to the transmission line 12, and supplied to the speech synthesizer shown in FIG.

In the speech synthesizer, the demultiplexer 11 demultiplexes and decodes the multiplexed data based on the input analysis information.

The decoded LPG coefficient data a is passed through the speech synthesis filter 1
3, the pitch period information C is the pitch, {Jless generator 14
Then, the voiced/unvoiced/silent information b is supplied to the switch 15, and the short-time voice power data d is supplied to the variable amplifier 16. The speech synthesis filter 13 is configured as an all-pole digital filter of a predetermined order, and is configured as an all-pole digital filter of a predetermined order.
are used as coefficients of this filter.

The voiced/unvoiced/silent information b is supplied to the switch 15, and when this data specifies voiced, the output of the pitch pulse generator 14 is sent to the variable amplifier 16, and when it is unvoiced, the output of the noise generator 17 is sent to the variable amplifier 16. The switch 15 is switched to supply the signal to l6.

The noise generator 17 generates white noise, and this white noise is supplied to the variable amplifier 16 when the voiced/unvoiced/silence information b specifies whether it is voiceless or silent.

Pitch pulse generator l4 supplied with pitch period information C
generates a pitch pulse train of a frequency corresponding to this pitch period, further corrects the position of the aforementioned pulse train based on the instruction from the pitch pulse variable device 18, which is a feature of the present invention, and then supplies it to the switch 15.

The variable amplifier 16 performs weighted amplification on the input pitch pulse or white noise in accordance with the magnitude of the short-time voice power data d that is input separately, and then supplies this to the voice synthesis filter 13. Use as a driving sound source.

The speech synthesis filter 13 uses the input LPG coefficient data a as a filter coefficient, is driven by a driving sound source, reproduces a quantized synthesized waveform for each analysis frame, and converts the D/
Supplied to A converter l9.

The D/A converter 19 converts the input quantized composite waveform into an analog waveform, and converts it into an analog waveform.
s Filter) Send to 2Q. LPF20 is
A predetermined Takagi frequency cutoff filtering is performed, and the synthesized speech is sent to the output line 21.

By the way, as mentioned above, the conventional speech synthesizer has a pitch period, a pulse excitation point that drives the speech synthesis filter 13, and
In other words, excessive concentration of energy occurs periodically, resulting in synthesized speech that lacks naturalness.However, in this embodiment according to the present invention, the pulse excitation point is changed to the pitch pulse variable device 18.
Move more intentionally. As a result, excessive concentration of energy that occurs every fixed period (pitch period) is dispersed based on the pitch period. That is, the pitch period,
Alternatively, by intentionally moving the excitation point of the pulse a little earlier or a little later than that period, it is possible to temporally disperse the generation of excessive energy that conventionally occurred periodically, which improves the auditory sense. This eliminates the sense of discomfort and significantly improves the naturalness of synthesized speech.

Here, the pitch pulse variable device 18 uses a typical M sequence as a pseudo-random number, and when the lower two bits of the M sequence are (0.0) (1.1), the pulse pulse according to the pitch period is When it is (1.0), it is excited one sample (125 μSec) earlier than the pitch period, and when it is (1.0), it is excited one sample later. The contents of the pitch pulse variable device 18 are updated every sample.

〔Effect of the invention〕

As explained above, according to the present invention, the speech synthesizer is provided with a pitch variable device, and the pulse excitation point is determined at the pitch pulse period and its vicinity, so that excessive concentration of energy can be prevented and natural It is possible to prevent the sound from becoming a mechanically synthesized sound. In this way, by using the pitch pulse variable device that varies the energy excitation point, the energy can be dispersed over time, making it possible to generate synthetic speech with good naturalness and no auditory discomfort.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of a speech synthesizer of the present invention. 11... Demultiplexer, 12...
Transmission line, 13...Speech synthesis filter, 14...Pitch pulse generator, 15.Old...Switcher, 16...Variable amplifier, 17...・
Noise generator, 18...Pitch pulse variable device, 1
9...D/A converter, 20...L
P.F. Figure 1

Claims (1)

[Claims] Spectral envelope information indicating the macroscopic structure of the audio signal analyzed by a speech analyzer and sound source information indicating the fine structure of the audio signal are input, the spectral envelope information is driven by the sound source information, and a speech synthesis filter is used. A speech synthesizer that synthesizes speech includes a pitch pulse generator that generates a pitch pulse train of a predetermined frequency based on pitch period information of the sound source information, and a generation position of the pitch pulse train generated by this pitch pulse generator. a pitch pulse variable device for changing the pitch pulse, a noise generator for generating white noise, and selecting the output of the pitch pulse generator when the voiced/unvoiced/unvoiced information of the sound source information is voiced information; A speech synthesizer comprising: a switch that selects the output of the noise generator when the information is unvoiced/silent information; and a switch that drives the speech synthesis filter based on the output of the switch.