JPH01262598A - Utterance speed control circuit for voice synthesizing device - Google Patents

Abstract

PURPOSE:To generate a composition voice being natural and easily listenable to a listener even in case when an utterance speed has been varied by calculating a spectrum variation rate by analyzing an hourly variation of a spectrum and providing a function relation between the spectrum variation rate and the utterance speed. CONSTITUTION:The title circuit is provided with a spectrum variation analyzing means 4 for converting a digital sound signal which has been brought to A/D conversion by an A/D converting means 1, to a spectrum by an analyzing means 3, and calculating a spectrum variation rate by analyzing an hourly variation of its spectrum, and a function relation is provided between the spectrum variation rate and the utterance speed by an utterance speed control part 6. In such a way, in case of varying the utterance speed of a composite voice, a degree of shrinkage is varied in accordance with magnitude of a spectral variation, and a sound source and a composite frame period of a spectral parameter are determined, therefore, decay caused by time expansion and contraction of a phoneme whose spectral variation is violent can be prevented, and a voice being natural and easily listenable can be obtained.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention relates to a speech rate control circuit for a speech synthesizer, and in particular, even when the speech rate is changed in a speech synthesizer, it is possible to control the speech rate in a way that is natural to the listener. This invention relates to a speech rate control circuit that outputs synthesized sounds that are easy to hear.

[Prior art and problems to be solved by the invention] A conventional speech synthesis device is, for example, disclosed in Japanese Patent Application Laid-Open No.
As stated in the publication, the expansion and contraction ratio is reduced for plosive short consonants such as /p/, /l/, and /ni/.
In addition, for stable vowels, the expansion and contraction ratio was increased to control the speech rate of synthesized sounds on a phoneme-by-phoneme basis. However, simply changing the expansion/contraction rate on the phonetic side does not give sufficient consideration to the preservation of fine phonological properties, and has the disadvantage that an unnatural sound remains when listening to the synthesized sound.

Therefore, the main object of the present invention is to provide a speech rate control circuit that can generate synthesized speech that is natural and easy for listeners to hear even when the speech rate is changed in a speech synthesizer. .

[Means for Solving the Problem] The invention according to the first claim includes an input means for inputting audio, an A/D conversion means for converting the input audio into a digital signal, and an A/D conversion means for converting the input audio into a digital signal. analysis means for analyzing the converted digital audio signal and extracting feature parameters;
In a speech synthesis device comprising a synthetic speech converting means for converting extracted feature parameters into synthetic speech, and a D/A converting means for converting the converted synthetic speech into an analog signal and outputting it, A/D conversion is performed. A spectral change analysis means is provided to convert a digital audio signal into a spectrum, analyze temporal changes in the spectrum, and calculate a spectral change rate, and create a functional relationship between the spectral change rate and the speaking rate. This is a configuration of a speech rate control circuit.

The invention according to claim 2 includes: an input means for inputting a character string or a symbol string; a reading sequence converting means for converting the input character string or symbol string into a reading string; and a converted reading string. In a speech synthesizer equipped with synthetic speech conversion means for converting a sequence into synthetic speech and D/A conversion means for converting the converted synthetic speech into an analog signal, the converted pronunciation sequence is converted into a spectrum and its A spectral change analysis means for calculating a spectral change rate by analyzing temporal changes in the spectrum is provided, and a voicing rate control circuit is configured to provide a functional relationship between the spectral change rate and the utterance rate.

[Operation] The speech rate control circuit of the speech synthesizer according to the present invention converts a speech signal or a converted pronunciation sequence into a parameter related to a spectrum, calculates a temporal change in the spectral parameter, and responds to the temporal change. The synthesized audio signal is then time-stretched and output.

[Embodiments of the invention] FIG. 1 is a schematic block diagram of an embodiment of the present invention.

In FIG. 1, an audio input unit 10 is used to input uttered audio, and the uttered audio is input to the A/D
The signal is input to the converter 1 and converted into a digital signal using sampling signals at predetermined intervals. The audio converted into a digital signal is given to the spectrum analysis section 3,
It is converted into spectral parameters that carry phonological information. As this spectral parameter, for example, Saito
There are PARCOR coefficients as described in "Fundamentals of Speech Information Processing" by Nakata (published by Ohmsha).Please note that this PARCOR coefficient is an example of a spectral parameter, and this invention is particularly limited to this. It's not a thing.

The audio digital signal output from the A/D converter 1 is also input to the audio analyzer 2, and sound source parameters such as pitch period and power information, which are sound source information, are extracted from the audio digital signal. The sound source parameters extracted by the sound source analysis section 2 and the spectrum parameters analyzed by the spectrum analysis section 3 are provided to the synthesis section 5. The synthesis unit 5 generates synthesized speech using the spectral parameters and the sound source parameters.

That is, the synthesis unit 5 repeats the spectral parameters within a unit frame at intervals of the pitch period described above and converts them into a string of audio parameters. This synthesis section 5 includes, for example, J, D, Markeland A, H, Gray.
A speech waveform is synthesized using the above-mentioned spectral parameters and sound source parameters using a square multiplier lattice speech synthesis filter such as that described in "Linear Prediction of Speech" (published by Corona Publishing Co., Ltd.), written by John Jr. and translated by Suzuki (published by Corona Publishing). Note that the square multiplier lattice type voice synthesis filter is only an example, and other voice synthesis means may be used. The synthesized audio waveform is output as audio via the D/A converter 7.

Next, control of the speech rate of the synthesized speech mentioned above will be explained. The speech rate of the synthesized speech is controlled by the synthesis unit 5,
This is achieved by changing the synthesis frame period of the sound source pulses and spectrum parameters when synthesizing the speech waveform from the sound source parameters and spectrum parameters. The spectral change analysis section 4 and the speech rate control section 6 perform such vocal control. First, the spectrum change analysis section 4 uses the spectrum parameters calculated in the spectrum analysis section 3 to find the temporal change rate Δ. For example, Mt. Saga.

Itakura, “Personality information contained in dynamic measures of speech.

A dynamic scale using LPC cepstral regression coefficients as described in the Proceedings of the Acoustical Society of Japan (published in June 1974) can be used as the spectral change rate Δ. Of course, the scale of the spectral change rate is not limited to this.

FIG. 2(a) shows an example of a temporal change in the voice spectrum change rate Δ, and FIG. 2(b) shows the average value of this change rate Δ for each frame. Figure 2 (
c) is an example of the sound source pulse and spectral parameters when vocalizing at the same speed as the original voice, and Figure 2 (C)
, the sound sources of the i-th frame and the j-th frame are displayed. Also, the spectral parameters in each frame are vectors (P, .

P1□...p'= + , (P', , P'
2... It is expressed as p'ml.

Next, a case will be described in which the speaking speed of synthesized speech is reduced, that is, the speech is expanded in the time axis direction. In one embodiment of the present invention, as shown in FIG. 2(b), the time length of the sound source pulse is determined based on the spectral change rate Δ determined for each frame. In the example shown in FIG. 2(b), the spectral change rate Δi of the i-th frame is
), the new sound source pulse time length is Ni for the time length Li of the sound source pulse of the i-th original audio.
is determined by the following equation (1).

LN i =L i In this way, an inverse relationship is established between the magnitude of the value of the spectral change rate Δ and the expansion/contraction rate, and new sound source pulses and spectral parameters fP'+ and P' are obtained as shown in FIG. 2(d). 2...P-) is placed.

The above procedure can also be applied to increase the synthetic speech production speed, and the new sound source pulse time length LNi is determined by setting equation (1) to LNi-r, txΔi/Δmax-(2).

Here, the relationship between the spectral change rate and the expansion/contraction rate does not need to be an inverse correlation, and any function can be used.

FIG. 3 is a schematic block diagram showing another embodiment of the invention.

In FIG. 3, the input terminal 20 has an OCR (not shown)
Sentences, words, etc. expressed as character strings or symbol strings are input using manual means such as a computer or a keyboard. The input sentences and words are converted into a string of morphemes by the morphological analysis section 8 based on the contents stored in the morphological dictionary section 9. The morpheme dictionary unit 9 stores at least "pronunciation" and "part of speech," and uses, for example, Aizawa and Ehara, "Kana-Kanji Conversion by Computer," NHK Technical Research, 25-5, for input character strings or symbol strings. Division into morphemes is performed using a means such as the longest match method as described in 2003. This longest match method is an example of a means for morpheme division, and is not limited thereto.

The morphemes analyzed by the morphological analysis unit 8 are given to the pitch control processing unit 10, and based on the contents stored in the accent dictionary unit 11 and accent combination rule unit 12, pitch frequency components representing the pitch of the sound are determined. be done. The morpheme string to which the pitch frequency component has been added is provided to the speech parameter generation section 13, and is converted into a string of phoneme parameters based on the contents stored in the phoneme segment dictionary section 14. The phoneme segment dictionary section 14 holds phoneme segments, that is, phonemes that make up sentences and words, or phoneme chains such as CV and VC as parameters, and the order of phonemes that make up a morpheme or phonemes such as CV and VC. The phonemes are arranged according to the above pitch frequency, and the parameters within a unit frame are repeatedly converted into a string of voice parameters at intervals of the pitch frequency described above.

Below, the spectrum analysis section 15. Spectral change analysis section 1
61 synthesis section 170 speech rate control section 18 and D/A conversion section 19 perform the same operations as in the embodiment shown in FIG. 1 described above.

〔Effect of the invention〕

As described above, according to the present invention, when changing the speech rate of synthesized speech, the expansion/contraction rate is changed according to the magnitude of the spectral change, and the synthesis frame period of the sound source and spectral parameters is determined. Therefore, it is possible to prevent phonemes with severe spectral changes from collapsing due to time warping, and it is possible to obtain natural and easy-to-listen speech.

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram of an embodiment of the present invention. FIG. 2(a) is a diagram showing an example of the spectral change rate, and FIG. 2(b) is a diagram showing the spectral change rate shown in FIG. 2(a) averaged for each frame. Second
Figure (c) is a diagram showing the arrangement of sound source pulses and spectral parameters when the speed is the same as that of the original voice, and Figure 2 (d)
2(b) is a diagram showing the arrangement of new sound source pulses and spectral parameters determined based on the spectral change rate shown in FIG. 2(b). FIG. 3 is a schematic block diagram of another embodiment of the invention. In the figure, 1 is an A/D converter, 2 is a sound source analyzer, 3,
15 is a spectrum analysis section, 4.16 is a spectrum change analysis section, and 5. Nia is a synthesis unit, 6 and 18 are speech rate control units,
7 and 19 are D/A conversion units, 8 is a morphological analysis unit, 9 is a morphological dictionary unit, 10 is a pitch control processing unit, 11 is an accent dictionary unit, 12 is an accent combination rule unit, 13 is a speech parameter generation unit, 14 indicates the phoneme dictionary section. Patent applicant: A.T.A.R.T. 2-1 Co., Ltd.
- Proceedings Amendment (, filc) August 10, 1988 1, Case Description 1988 Patent Application No. 91131 Showa Year
Monthly Submission Characteristics 2, Name of the invention Speech rate control circuit for speech synthesis device 3, Relationship to the case of the person making the amendment Patent applicant address Kyoto Prefecture) ■Rakubu Seika-cho Oaza Inuiya Koza Sanpeidani 5 name Name: ATR Co., Ltd. Research Institute Representative: Eiji Yodoyo 4; Agent Address: Sumitomo Bank Minamimorimachi Building 6, 2-1-29 Minamimorimachi, Kita-ku, Osaka; Drawing subject to amendment 7; Contents of the amendment Figure 2 of the drawings will be amended as shown in the attached sheet. that's all

Claims (2)

[Claims] (1) An input means for inputting audio, an A/D conversion means for converting the audio input from the input means into a digital signal, and a digital audio signal converted by the A/D conversion means. an analysis means for analyzing and extracting feature parameters; a synthetic speech conversion means for converting the feature parameters extracted by the analysis means into synthetic speech; and a synthetic speech conversion means for converting the synthetic speech converted by the synthetic speech conversion means into an analog signal. A speech synthesis device comprising a D/A conversion means for outputting a signal, the digital audio signal converted by the A/D conversion means is converted into a spectrum, and temporal changes in the spectrum are analyzed to determine the rate of change in the spectrum. A speech rate control circuit, comprising a spectral change analysis means for calculating, and creating a functional relationship between the spectral change rate and the speech rate. (2) an input means for inputting a character string or a symbol string; a pronunciation conversion means for converting the character string or symbol string input by the input means into a pronunciation; and conversion by the pronunciation conversion means. A speech synthesis device comprising a synthetic speech conversion means for converting the read sequence into a synthesized speech, and a D/A conversion means for converting the synthesized speech converted by the synthetic speech conversion means into an analog signal, spectral change analysis means for converting the pronunciation sequence converted by the conversion means into a spectrum, analyzing temporal changes in the spectrum to calculate a spectral change rate, and determining a functional relationship between the spectral change rate and the speaking rate. A speech rate control circuit characterized by having: