JPH0421900A - Sound synthesizer - Google Patents

Abstract

PURPOSE:To output a synthetic sound having high naturality by providing the sound synthesizer with a fluctuation control part for applying fluctuation to the amplitude, frequency and phase information of a sound source signal and a voice route filter part for simulating human's voice route, nasal cavity and oral cavity. CONSTITUTION:In the case of synthesizing voices, a sound source reproducing part 2 extracts a necessary sound source signal corresponding to one period or plural periods from a sound source storing part 1 storing sound sources for plural periods extracted from natural sounds and repeatedly sends the waveform to the succeeding reference frequency control part 3, which converts the inputted sound source wave into a required reference frequency. The fluctuation control part 4 applies random amplitude modulation, phase modulation, frequency modulation, etc., to he sound source signal inputted from the control part 3 to add fluctuation included in natural sounds to the inputted sound. The output of the control part 4 is inputted to the voice route filter part 5, which outputs a synthetic sound adopting a Formant synthesizing system e.g. for simulating human's voice route, oral cavity and nasal cavity. Consequently, a synthetic sound having extremely high naturality can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a speech synthesis device for converting text into speech.

Conventional technology A variety of speech synthesis methods have been devised at present, including LSP, P
These include ARCOR, cepstral force method, and terminal analog formant synthesis method. Among these, the formant synthesis method has the best physical correspondence with the human vocal mechanism and has been used for a long time.

FIG. 6 shows the configuration of a conventional formant speech synthesis system. Here, for simplicity, only vowel synthesis will be considered. In the figure, 11 is a sound source wave generation section, and 12 is a pomanto synthesis section consisting of a resonator and an anti-resonator. Ma Tsui
A sound source wave generating section 1 synthesizes a sound source wave having a desired fundamental frequency. Currently used sound source wave models ζ As shown in Figure 6, continuous pulse processing or models with shapes that imitate natural sound source waves are used. This sound source signal is manually inputted to the formant synthesis section 12, imparted with vocal tract resonance characteristics, and outputted as a synthesized sound 2.

With this method, there is very little sound source information that can be handled in a sound source signal such as 1 yaw 4 2 yaw, and it is not possible to expect more than a certain degree of naturalness. Natural sound source signal (no individuality)
It is believed that it contains information that is extremely important for speech, such as age. However, in the conventional sound source model as described above, important information is largely missing, and the reality is that the output from the synthesizer becomes a mechanical sound.

Problems to be Solved by the Invention The present invention takes into consideration the above-mentioned drawbacks of the conventional example and provides a technology that can output highly natural synthesized sounds.

Means for Solving the Problems A sound source signal storage unit stores a part of extracted sound source signals generated in human vocal cords, and a sound source signal of an arbitrary length is reproduced using the sound source signal of the sound source signal storage unit. a fundamental frequency control section that controls the fundamental frequency of the sound source signal that is the output of the sound source signal reproduction section, and an amplitude IN frequency of the sound source signal that is the output of the fundamental frequency control section;
The configuration includes a fluctuation control section that adds fluctuation to phase information, and a vocal tract filter section that simulates the human vocal tract, oral cavity, and nasal cavity.

By using the above-mentioned means, speech synthesis processing can be performed without losing the detailed information of the natural sound source, and high-quality speech can be obtained.

Embodiment FIG. 1 shows the configuration of a speech synthesis apparatus in an embodiment of the present invention, which aims at high-quality speech synthesis processing without losing the detailed information of natural sound sources.

In the figure, 1 is a sound source storage section, 2 is a sound source reproduction section 3 is a fundamental frequency control section, 4 is a fluctuation control section, and 5 is a vocal tract filter section.

The operation of the speech synthesis apparatus of this embodiment configured as described above will be described below. Here, for simplicity, only vowel synthesis will be considered.

The sound source storage unit 1 stores, for example, one cycle or a plurality of cycles of the fundamental frequency of a sound source signal obtained by removing vocal tract resonance characteristics from natural speech. When performing speech synthesis, first, the sound source reproduction section 2 extracts a necessary one period or a plurality of periods of the sound source signal from the sound source storage section 1, repeats its waveform, and sends it to the next fundamental frequency control section 3. At this time, the front end and rear end of the sound source signal are smoothly connected. An example of this connection method will be explained. Figure 2 (Dai) shows an example of this waveform connection.The sound source waveform is digitized and set to S[i] (i=1.2., , , N), and the following distance scale D(k) is defined. do.

Calculate the value k that minimizes this D(k) t,, i=o
Connect the points of and i=. Moreover, this distance ((k) Σ S(k+j,)-8(i) D(k) =-Σ(S(k+1)-8(i)), etc. can also be considered.

Next, the fundamental frequency control section 3 converts the sound source wave output from the sound source reproduction section 2 into a desired fundamental frequency.

In this conversion method, for example, a sound source signal is digitized and linear interpolation is performed between each sample point. FIG. 3 shows this linear interpolation.

Next, the fluctuation control section 4 performs, for example, random amplitude transformation, phase change temperature, frequency modulation, etc. on the sound source signal output from the fundamental frequency control section 3 to add fluctuations inherent in natural speech. The output from the fluctuation control section 4 is input to the vocal tract filter section 5. Here, a formant synthesis method will be considered as an example of the vocal tract filter section 5. FIG. 4 shows how the vocal tract filter section 5 performs signal processing using formant synthesis. Here, resonators corresponding to each formant frequency of a desired phoneme change the spectrum of the sound source signal and output it as a synthesized sound.

As described above, according to this embodiment, it is possible to effectively utilize natural sound sources and output high-quality synthesized sounds.

Next (see Figure 5), we will now explain a speech synthesis device in a second embodiment of the present invention, which aims to perform high-quality synthesis with arbitrary sound quality. For simplicity, only vowel synthesis will be considered.

In the figure, 6 is a phoneme recognition unit that recognizes each phoneme of the input audio signal, 7 is a whitening filter, 8 is a vocal tract parameter extraction R, and 9 is a vocal tract parameter storage unit 10 is a sound source information extraction unit.

First, when the operator registers the desired sound quality in this synthesizer,
The operator inputs natural speech of that quality to the phoneme recognition unit 6. At this time, it is convenient to input a predetermined vowel sequence. The phoneme recognition unit 6 recognizes each vowel and extracts information such as boundary positions between phonemes from the input speech.

On the other hand, the same signal input to the phoneme recognition section 6 is input to the whitening filter section 7. Whitening filter section 7 (Friend) At the same time as performing operations to remove resonance characteristics of the vocal tract, information on vocal tract parameters such as formant frequencies is also calculated. The case of canceling the resonance characteristics of each formant is shown. From the output of the whitening filter section 7 and the output of the phoneme recognition section 6, the vocal tract parameter extraction section "8 detects the position of the stationary part of each vowel. Parameters related to the vocal tract, such as the frequency and bandwidth of each formant, are extracted and stored in the vocal tract parameter storage section 9.The sound source information extraction section 10 extracts the signal from which the vocal tract characteristics have been removed, which is the output of the whitening filter section 7. Of these, the vowel stationary portion is extracted as a representative sound source waveform and stored in the sound source storage unit 1.

This completes the preparation for synthesizing speech with the desired quality. The subsequent synthesis process is the same as the process explained in FIG. By using the information corresponding to the specific voice quality, a synthesized voice of high quality and close to the desired voice quality can be obtained.

As described above, according to this embodiment, by analyzing the voice of a specific person and storing its characteristics, it is possible to synthesize a desired voice quality with high quality using natural sound sources.

Effects of the Invention As described above, according to the present invention, extremely natural synthesized sounds can be obtained by repeatedly using sound source waves extracted from natural speech instead of pulses or simple models as sound sources.

Furthermore, since sound source waves greatly contribute to the individuality of speech, high-quality synthesis of any sound quality can be achieved by attaching sound source wave extraction means, vocal tract parameter extraction means, etc. to this synthesis device. The practical effects of this are significant.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of a speech synthesis device according to a first embodiment of the present invention. Fig. 2 is a block diagram showing how the sound source signals are connected in this embodiment. Fig. 3 is a diagram showing the fundamental frequency of the sound source signal in this embodiment. FIG. 4 is a block diagram showing an example of the operation of the vocal tract filter section. 1...Sound source storage chant 2...Sound source playback section, 3...Fundamental frequency control section 4...Fluctuation control section 5...Vocal tract filter section 6...Phoneme recognition section 7...White color filter eye 8... Vocal tract parameter extraction ability 9... Vocal tract parameter storage sleep 10... Sound source information extraction ability 11...
Sound source wave generation starvation 12... Formant synthesis accent

Claims (4)

[Claims] (1) A sound source signal storage section that stores a part of extracted sound source signals generated in the human vocal cords, and a sound source signal reproduction that reproduces a sound source signal of arbitrary length using the sound source signal of the sound source signal storage section. a fundamental frequency control section that controls the fundamental frequency of the sound source signal that is the output of the sound source signal reproduction section; and a fluctuation that imparts fluctuations to the amplitude, frequency, and phase information of the sound source signal that is the output of the fundamental frequency control section. A speech synthesis device comprising a control section and a vocal tract filter section that simulates a human vocal tract, oral cavity, and nasal cavity. (2) Claim 1 characterized in that the stored sound source signal is one cycle or multiple cycles of the fundamental frequency, and includes an audio signal reproducing unit that repeats the signal to produce a long-time sound source signal. The speech synthesizer described in . (3) The speech synthesis device according to claim 1, further comprising a fundamental frequency control section that performs linear interpolation between sample points when digitizing the sound source signal and changing the fundamental frequency. (4) a phoneme recognition unit that recognizes phonemes in the input audio signal; a whitening filter unit that acts to cancel the resonance characteristics of the vocal tract from the input audio signal; and a whitening filter unit that is connected to the whitening filter unit and connected to the phoneme recognition unit. a vocal tract parameter extraction unit that extracts vocal tract parameters of a vowel stationary part based on the recognition result obtained by the phoneme recognition unit from the vocal tract parameter series obtained from the connected whitening filter unit; a vocal tract parameter storage section that stores a vocal tract parameter series such as each vowel output from the extraction section; a sound source information extraction section that is connected to the whitening filter section and extracts sound source information such as a sound source waveform and fundamental frequency; a sound source information storage unit that stores sound source information that is the output of the sound source information extraction unit; sound source information that is output from the sound source information storage unit during speech synthesis; and vocal tract parameter information that is output from the vocal tract parameter storage unit; 2. The speech synthesis apparatus according to claim 1, wherein the speech synthesis process is performed using the speech synthesis apparatus.