JPH02244100A - Noise sound source signal forming device - Google Patents

Abstract

PURPOSE:To obtain high-quality synthesized voices by mixing phase equalized residual waveforms and random-phased residual waveforms. CONSTITUTION:The phases of the power spectra of the residual waveforms obt. by executing reverse filtering of man's original voice signals by a reverse filter 4 are unified by a phase equalizing circuit 6. The residual waveforms randomized in the phase by a random phasing circuit 6 are added by a residual synthesizing circuit 7 to the residual waveform unified in the phases, by which the noise signals superposed on the period signals in the state of preserving the spectra of the residual waveforms are simulated and the spectra are preserved with respect to a change in pitch period. The high-quality voices are synthesized in this way.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a drive sound source signal generation device for driving a circuit that simulates the spectral envelope of speech in a speech synthesis device.

[Prior art and problems to be solved by the invention] Conventional speech synthesis devices include a driving sound source signal generating device that expresses the pitch period, amplitude, and spectral fine structure of speech, and a driving sound source signal generated by this device. and a vocal tract filter for simulating the spectral envelope of speech.

The vocal tract filter is P A RCOR (partial a
(auto-corralation) method and LSP
(linear pragmatic coding)
It consists of a digital filter such as a method, and the driving sound source is switched between impulse and white noise.

As the driving sound source, a residual waveform obtained by inverse filtering an original human voice signal may be used. This is aimed at improving the sound quality by covering the components that cannot be approximated by the spectrum parameters with the sound source.

However, 1. In regular regular speech synthesis, the pitch period at the time of synthesis is different from that at the time of analysis, and the residual waveform is used as it is to synthesize speech with a short pitch period.

The residual waveform is partially deleted and the spectrum is not saved.

Therefore, in order to concentrate the power of the residual waveform in a short time while preserving the spectrum of the residual waveform, the time waveform that has been inversely Fourier transformed with all the phases of the power spectrum of the residual waveform zeroed is There is a method for using it as a sound source signal. However, with this method, it is not possible to obtain synthesized speech that includes the effects of noise caused by narrowing of articulation and noise generated near the vocal cords, which is said to be unique to a single female voice.

An object of the present invention is to solve these problems and thereby obtain "natural synthesized speech that is closer to human speech."

Note that there are many documents regarding speech synthesis.

For example, the monthly magazine [Information Processing J (vol. 24. N
o8°Aug, 1983) pp. 993-100
In the commentary "Speech Synthesis Technology" published in 2003, a method is described in which a pulse period and noise are switched as a sound source as a speech synthesis model. Also, JP-A-58-88798
The publication describes a speech synthesis method in which a waveform with a period of about 1 pitch cut out from a residual signal is used as a driving waveform.

[Means and effects for solving the problem] The present invention adds a residual waveform whose phase is randomized to a residual waveform whose phase is aligned to 0, so that it overlaps the periodic signal while preserving the spectrum of the residual waveform. This method simulates a noise signal that has been generated by a noise signal, preserves the spectrum even when the pitch period changes, and thereby enables high-quality speech synthesis.

〔Example〕

Hereinafter, an embodiment of the present invention will be described using FIG. 2.

FIG. 1 is a block diagram showing an embodiment of a speech synthesis device according to the present invention. This speech synthesizer is.

The sound source signal generated by the drive sound source signal generation device 1 is
The vocal tract filter 3 is driven by a signal whose amplitude is controlled by the amplitude control circuit 2.

The vocal tract filter 3 simulates the spectral characteristics of speech, and specifically uses an LSP synthesis circuit or the like.

The drive sound source signal generation device 1 aligns the phase of the power spectrum of the residual waveform with the inverse filter 4 of the vocal tract filter 3 to obtain a residual waveform by performing inverse filtering of the original human voice A symbol. A random phasing circuit 6 randomizes the phase of the power spectrum of the residual waveform, and a residual combining circuit 7.

Specifically, the inverse filter 4 performs LSP analysis on the original audio signal and outputs a residual waveform.The zero-phase equalization circuit 5 is for temporal power concentration of the residual waveform. ,
On the other hand, the random phasing circuit 6 is used to spread the temporal power of the residual waveform.

FIG. 2 is an explanatory diagram of phase equalization processing by the phase equalization circuit 5. First, a windowing process such as a Hamming window is applied to the residual waveform obtained by the inverse filter 4 as a discrete window function of a constant length. Next, the spectrum of the residual waveform is obtained by Fourier transformation.

Next, after setting the phase of each frequency of the spectrum to zero, it is returned to a time waveform by inverse Fourier transform.

In general, the phase may be set to a certain constant value instead of zero, but it is inevitable that the power will be diffused (the concentration of power will be lower) than when the phase is set to zero.

FIG. 3 is an explanatory diagram of random phasing processing by the random phasing circuit 6. First, the residual waveform obtained by the inverse filter 4 is subjected to windowing, such as a Hamming window, and then subjected to Fourier transformation. Next, after randomizing the phase of the spectrum, it is returned to a time waveform by inverse Fourier transformation.

FIG. 4 shows examples of the original residual waveform obtained by the inverse filter 4, the residual waveform whose phase has been equalized (zero phase), and the residual waveform whose phase has been changed to random.

FIG. 5 shows an example of the residual synthesis circuit 7. In this example, the adder circuit 10 stores in the buffer 11 a time waveform obtained by adding a phase equalized residual waveform and a randomly phased residual waveform each having a length of about one pitch period at a certain ratio. I'll keep it. Then, the repetitive waveform generating section 12 generates a composite residual waveform (drive sound source signal) by repeatedly reading out the waveform in the buffer 11 at a designated pitch period.

Here, since the residual waveform differs depending on the type of phoneme and the part of the phoneme, it is necessary to prepare a plurality of residual waveforms and update the residual waveform as appropriate to obtain a composite residual waveform. This update cycle is changed in units of phonemes and frames.

FIG. 6 shows another example of the residual synthesis circuit 7. In this example,
The phase equalized residual waveform and the random phased residual waveform are stored in separate buffers 14.15. The randomly phased residual waveform buffer 15 is divided into a plurality of banks. The phase equalization residual waveform in the buffer 14 is generated by the repetitive waveform generator 1
6 is read out repeatedly at pitch intervals. boost,
In the buffer 15 for the random phased residual waveform, the bangs are randomly selected by the bank switching circuit 18 according to the random numbers generated by the random number generation circuit 17, so that the divided portions of the random phased residual waveform in the buffer 15 are randomly selected. It is taken out.

The waveform taken out from the buffer 14.15 is subjected to amplitude control by the amplitude control circuit 19.20, and then sent to the addition circuit 21.
By adding (mixing) , a composite residual waveform is obtained.

That is, in this embodiment, the randomly phased residual waveform is mixed with the phase equalized residual waveform as a random repeat, rather than a monotonous repeat.

The mixing ratio of these two residual waveforms is determined by the amplitude control circuit 19.
It is determined by the amplitude control value (yw y-1) at 20.

Although not shown, according to another embodiment of the present invention, the mixing ratio of the phase equalized residual waveform and the random phased residual waveform is changed depending on the type of phoneme and the frame position.

More specifically, a mixing ratio is determined in advance for each phoneme or frame and stored in memory, and in the configuration shown in FIG. 5 or 6, for example, the mixing ratio is selectively read out from the memory. By controlling the adder circuit 10 or the amplitude control circuits 19 and 20 in the following manner, voiced sounds and unvoiced sounds can be handled uniformly without distinction.

According to yet another embodiment of the present invention, the mixing ratio of the phase equalized residual waveform and the random phased residual waveform is changed depending on the slope of the spectral envelope.

〔Effect of the invention〕

As explained above, according to the present invention, since the residual waveform that has been equalized by one phase and the residual waveform that has been randomly phased are mixed,
Since it is possible to simulate the noise component superimposed on the periodic signal while preserving the spectrum of the residual waveform, and to preserve the spectrum even when the pitch period changes, it is possible to obtain higher quality synthesized speech. Furthermore, by changing the mixing ratio of the phase-equalized residual waveform and the random-phased residual waveform depending on the type of phoneme and the frame position, it is possible to uniformly handle voiced sounds and unvoiced sounds without distinguishing between them.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing an embodiment of the speech synthesis device according to the present invention, Fig. 2 is an explanatory diagram of phase equalization processing, Fig. 3 is an explanatory diagram of random phasing processing, and Fig. 4 is an illustration of the original residual. A waveform diagram showing an example of a difference waveform, a waveform after phase equalization, and a waveform after random phasing, FIG. 5 is a block diagram showing an example of a residual synthesis circuit,
FIG. 6 is a block diagram showing another example of the residual synthesis circuit. DESCRIPTION OF SYMBOLS 1... Drive sound source signal generation device, 2.19.20... Amplitude control circuit, 3... Vocal tract filter, 4... Inverse filter, 5...
・Phase equalization circuit, 6...Random phase equalization circuit, 7.
... Residual synthesis circuit, 10.11 ... Addition circuit, 11
．． 14, 15...Buffer, 12.16...Repetitive waveform generator. 17...Random number generation circuit, 18...Bank switching circuit. Figure 2 Figure 3 Figure 4

Claims (3)

[Claims] (1) means for obtaining a residual waveform by inverse filtering an audio signal; and means for equalizing the phase and randomizing the power spectrum of the residual waveform obtained by the means;
means for mixing the phase-equalized residual waveform and the random-phased residual waveform at a predetermined ratio; means for storing the temporal waveform mixed by the means; 1. A drive sound source signal generation device for a speech synthesis device, comprising means for repeatedly extracting a time waveform. (2) means for obtaining a residual waveform by inverse filtering the audio signal, and means for phase equalizing and random phasing of the power spectrum of the residual waveform obtained by the means;
means for separately storing the phase-equalized residual waveform and the randomly phased residual waveform by the means; and a means for periodically and repeatedly retrieving the phase-equalized residual waveform, 1. A driving sound source signal generating device for a speech synthesis device, comprising means for randomly extracting divided portions of a randomly phased residual waveform and mixing them at a predetermined ratio. (3) The mixing ratio of the phase-equalized residual waveform and the random-phased residual waveform is determined by the phoneme type, frame position,
The driving sound source signal generating device according to claim 1 or 2, wherein the driving sound source signal generating device changes depending on the slope of the spectrum envelope.