JPH02247700A - Voice synthesizing device - Google Patents

Abstract

PURPOSE:To stabilize synthesized voices and to improve quality thereof by determining the spectral parameter of an original voice signal which is subjected to a high-band stress and passing the original voice signal through a reverse filter of this spectrum. CONSTITUTION:The original voice is subjected to the high-band stress in a high-pass filter 1 and is subjected to the spectral analysis of the original voice at every frame in an LSP analysis circuit 2, from which the voice is outputted as an LSP parameter. The original voice obtd. by passing the LSP reverse filter 3 having the characteristics reverse from the spectral characteristics by the LSP parameter is subjected to a high-band suppression and is made into a residual waveform. This residual waveform is edited in accordance with the data on pitch and amplitude by a residual waveform synthesis circuit 6, by which the synthesized residual waveform is formed and is outputted as the synthesized voice by an LSP synthesis filter 4. The residual waveform of the large power is thereby obtd. and the high-quality synthesized voice is stably obtd.

Description

[Detailed Description of the Invention] The present invention relates to a speech synthesis device, and more particularly, to a regular speech synthesis device using a residual waveform, and relates to an analysis and synthesis system for synthesizing high-quality speech. .

A conventional speech synthesizer consists of a vocal tract filter that expresses the spectral envelope of speech, and a driving sound source signal generation device that expresses the pitch period, amplitude, and spectral fine structure of speech. Usually, the vocal tract filter is composed of a digital filter such as the PARCOR method or the LSP method, and the driving sound source is used by switching between impulse and white noise (Hide Shiratori: Speech Synthesis Technology, Information Processing Vol. 1.24 No. 8)
pp. 993-1000, the sound source is switched between pulse period and noise as a speech generation model).

As a driving sound source, a residual waveform obtained by inverse filtering the original human voice signal may be used (Tokuko Kokō 5).
In Japanese Patent No. 8-88798, a waveform with a period of about 1 pitch extracted from the residual signal is used as a cantering waveform). This aims to improve sound quality by covering components that cannot be approximated by spectral parameters with the sound source.

However, in the conventional residual waveform, the power of the residual waveform may be extremely low depending on the phoneme, which causes oscillation of the synthesis filter.

■---The present invention was made in view of the above-mentioned circumstances.
In particular, this was done with the aim of increasing the power of the driving sound source by using a residual waveform with suppressed high frequencies, thereby reducing the oscillation of the synthesizer and obtaining a more stable and high-quality synthesized sound. be.

In order to achieve the above-mentioned object, the present invention provides a speech synthesis device that includes a circuit that simulates the characteristics of the vocal tract or the spectral envelope of speech and a sound source that drives the circuit. It has an emphasizing filter, an analysis circuit that obtains parameters representing the spectral envelope of the input signal, and an inverse filter for the frequency characteristics of this spectral envelope. Drive using a residual waveform with suppressed high frequencies obtained by passing the original speech signal through an inverse filter, or a circuit that simulates the characteristics of the vocal tract or the spectral envelope of speech and a sound source that drives the circuit. A speech synthesis device comprising: an adaptive inverse filter for removing the spectral slope of the original speech signal; an analysis circuit for obtaining parameters expressing the spectral envelope of the input signal; and an inverse filter for the frequency characteristics of the spectral envelope; After removing the slope of the original audio signal spectrum, the spectrum parameters are obtained using an analysis circuit, and the JM sweat voice signal is applied to an inverse filter of this spectrum.
This method is characterized in that the spectral slope of the original audio signal is included in the sound source signal, and conversely, the spectral slope is removed from the spectral parameters. Hereinafter, the present invention will be explained based on examples.

Accordingly, in the present invention, the spectral parameters of the original audio signal with high frequencies emphasized are obtained, and the residual waveform with the high frequencies suppressed is obtained by passing the original audio signal through an inverse filter of this spectrum. changes from the conventional impulse-like shape to a triangular wave-like one, and the power becomes relatively large.As a result, the gain of the synthesis filter decreases, which reduces the oscillation of the synthesizer and makes it more stable and high-frequency. It is possible to obtain high-quality synthesized speech.

FIG. 1 is a block diagram for explaining an embodiment of the invention described in Item 1 of the claims.In the figure, 1 is a high-pass filter, 2 is an LSP analysis circuit, 3 is an LSP inverse filter, and 4 is an LSP synthesis filter, 5 is a parameter and residual waveform storage unit, 6 is a residual synthesis circuit, the original voice is audio material for creating 1 synthesis parameter, and the original voice is processed in high frequency by high frequency filter 1. After being highlighted, LSP analysis time lf! Sent to t2.

The LSP analysis circuit 2 performs spectrum analysis of the high-frequency emphasized original speech for each frame and outputs it as an LSP parameter representing the spectrum envelope. Also, the original waveform is L
It is passed through an SP inverse filter 3. Here, the LSP inverse filter 3 is a filter having spectral characteristics opposite to those given by the LSP parameters. The spectral envelope indicated by this LSP parameter emphasizes the high range compared to that of the original voice, so
The output of the SP inverse filter 3 becomes a residual waveform with high frequencies suppressed. FIG. 2 shows the original audio signal (a), the residual waveform without high frequency suppression (b), and the residual waveform with high frequency suppression (CQ).

At the time of synthesis, the high-frequency suppressed residual waveform is edited by the residual synthesis circuit 6 based on pitch and amplitude data to become a composite residual waveform. The synthesized residual waveform is output as synthesized speech by LSP synthesis [t4.

In addition to the LSP parameters of this embodiment, parameters representing other spectral characteristics such as Percoll, LPG, and formant can also be used.

FIG. 3 is a block diagram for explaining one embodiment of the invention as set forth in claim 2. In the figure, 10 indicates an adaptive filter, and the adaptive filter 10 is configured to calculate the first-order correlation in the original speech. The number is calculated by a correlator, the slope of the spectrum is removed from the spectrum parameter, and the original speech signal is passed through an inverse filter of this spectrum to obtain a residual waveform with large power. First, the slope of the spectrum is corrected to make the spectrum flat. Here, if the first-order correlation coefficient is r, this can be achieved by a programmable filter with the characteristic F(z) = 1-rLz-1. .

The operations of the LSP analysis circuit 2 to residual synthesis circuit 6 are as follows.
Since this is exactly the same as the case shown in FIG. 1, detailed explanation thereof will be omitted.

As is clear from the above description, according to the invention of claim 1, the spectrum analysis is performed with the high frequency band emphasized, and the spectral parameters of the original audio signal with the high frequency band emphasized are determined. Since the original speech signal is passed through this spectral inverse filter, a residual waveform with large power is obtained, which reduces the oscillation of the synthesizer, making it possible to obtain a more stable and high-quality synthesized speech. Additionally, a high-frequency emphasis filter is used to correct the spectral tilt characteristic of voiced sounds, flattening the spectrum and increasing the accuracy of high-frequency analysis.

It also has the effect of suppressing the phenomenon in which low-order formants appear as an apparently very sharp Q.

Further, according to the invention of claim 2, since the slope of the spectrum is removed from the spectrum parameter by the adaptive filter and the original audio signal is passed through the inverse filter of this spectrum, a residual waveform with large power can be obtained. This reduces synthesizer oscillations, resulting in more stable and high quality synthesized speech. In addition, voiced sounds 1,!, by adaptive filters. Corrects the slope of the spectrum regardless of the voice,
By flattening the spectrum, high-frequency analysis accuracy is improved. It also has the effect of suppressing the phenomenon in which low-order formants appear as an apparently very sharp Q.

[Brief explanation of drawings]

FIG. 1 is a block diagram for explaining one embodiment of the invention as set forth in claim 1, and FIG. 2 shows a residual waveform (b) without high frequency suppression and a residual waveform (b) with high frequency suppression. Diagram showing the difference waveform, 3rd
The figure is a configuration diagram for explaining an embodiment of the invention set forth in claim 2. 1...High-pass filter, 2...LSP analysis circuit, 3.
...LSP inverse filter, 4...LSP synthesis filter,
5... Parameter, residual waveform storage section, 6... Residual synthesis circuit, 10... Adaptive filter. District 1, Figure 2

Claims (1)

[Claims] 1. A speech synthesis device comprising a circuit that simulates the characteristics of the vocal tract or the spectral envelope of speech and a sound source that drives the circuit, which includes a filter that emphasizes the high frequency range of the original speech signal, and a filter that emphasizes the high frequency range of the input signal. It has an analysis circuit that obtains parameters expressing the spectral envelope, and an inverse filter for the frequency characteristics of this spectral envelope, obtains the spectral parameters of the high-frequency emphasized original audio signal, and passes the original audio signal through this spectral inverse filter. A speech synthesis device characterized in that it is driven using a residual waveform obtained by suppressing high frequencies. 2. In a speech synthesis device consisting of a circuit that simulates the characteristics of the vocal tract or the spectral envelope of speech and a sound source that drives the circuit, an adaptive inverse filter that removes the spectral slope of the original speech signal and the spectral envelope of the input signal are used. , and an inverse filter for the frequency characteristics of this spectral envelope. After removing the slope of the original audio signal spectrum, the analysis circuit calculates the spectral parameters, and the original audio signal is converted into this spectrum. 1. A speech synthesis device characterized in that a spectral slope in an original audio signal is included in a sound source signal and conversely removed from a spectral parameter by applying the spectral slope to an inverse filter.