JPH02106799A - Synthetic voice emotion imparting circuit - Google Patents

Abstract

PURPOSE:To obtain the voice having natural emotion and to control the intensity thereof by controlling only the three elements of rhythms; basic frequency, time structure and amplitude among extracted characteristic parameters. CONSTITUTION:The emotion imparting circuit 4 subjects the basic frequency, time structure and amplitude parameters among the sound source parameters outputted from a sound source analyzing section 2 to function conversion and synthesizes the voice waveforms by these sound source parameters and spectral parameters in a synthesizing section 5. Namely, the fresh basic frequency f'(t) and the fresh amplitude parameter P'(t) are determined with respect to the basic frequency f0(t) among the sound source parameters outputted from the sound source analyzing section 2. The converted fresh sound source parameters are sent together with the spectral parameters to the synthesizing section 5 from which the parameters are outputted in the form of the synthesized voice waveforms via a D/A conversion section 6. The synthesized sounds having the 'emotion and feel' which are natural and are comfortable to be listened by listeners are outputted in this way.

Description

[Detailed Description of the Invention] U Industrial Field of Application This invention relates to a nine-circuit with synthesized speech emotion, particularly in a speech synthesizer such as an analysis synthesizer or a rule synthesizer, which is natural and easy to hear for the listener. Synthetic voice emotion that outputs a synthesized voice with "emotions/feelings"
related to the road.

[Prior art and problems to be solved by the invention] As an example of a conventional speech synthesis device, for example, Japanese Patent Laid-Open No. 62-1
``Speech rule synthesis horn type'' as described in Japanese Patent No. 38898 is known. This phonetic rule synthesis h° formula consists of 11 questions and commands.

Kinichi Sentence 1 Based on the sentence type information such as desire sentences and inspirational sentences, the corresponding intonation is generated to generate a fundamental frequency time change pattern described by the response function of a critical damped quadratic linear system. It is.

However, in such a speech rule synthesis method,
By simply changing the fundamental frequency pattern, it is not possible to fully express the detailed emotional expressions that human voices have, but when listening to the synthesized sound, it still sounds unnatural.

Therefore, the main object of the present invention is to
! 7, it is an object of the present invention to provide a synthesized speech emotion fζj Lee circuit that can generate synthetic speech having a natural emotion of "anger."

[Means for solving the problem] The invention according to claim 12iI is a digital BJ”j
an input means for inputting f+j, an analysis means for analyzing the input digital audio signal and extracting feature parameters, and a synthetic Ishikawa transformation for converting the feature parameters extracted by the analysis means into a synthesized sound j4i. Means and converted synthesis 8f! In a speech synthesizer equipped with a D/A conversion means for converting 1 to an analog signal No. 2 and outputting it, the fundamental frequency 1 time) 1η structure and amplitude parameter of the characteristic parameters extracted by the analysis means are functioned. 7 circuits are constructed so as to convert the synthesized voice emotion tI and generate a message expressing the emotion of 1 anger by means of the synthesized amount tI conversion means.

Second. The invention related to the information item includes an input means for inputting a character string or a notation, a reading sequence conversion means for converting the input character string or symbol string into a reading string, and a converted reading string. Synthetic speech converting means for converting the string into synthetic speech, and D/A converting the converted synthetic speech into analog signals. 11. The fundamental frequency of the characteristic parameters in the Ti synthesis device with 6jk additional conversion means. The synthesized sound per emotion ('J'' is composed of 7 circuits so as to generate a synthetic stone expressing the emotion of "anger" by functionally converting the ff interval and amplitude parameters, and using the synthesized sound and ji conversion means. be.

[Made in Japan] The emotion-111 circuit according to the present invention converts an audio signal or a converted pronunciation sequence into parameters related to a spectrum, while extracting or extracting parameters related to a sound source, and extracts or extracts parameters related to a sound source. The basic frequency is output for 1 hour (after controlling the IXi structure and amplitude parameters).

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

First, with reference to FIG. 1, the structure and operation of an embodiment of the present invention will be described. Digimol sound 11 signal is input to the digital audio signal input section], and the digital 8F! 1 signal is sent to the sound source analysis section 2 and the spectrum analysis section 3
I can get JJ. The spectrum analysis section 3 converts the digital voice number 11 into spectral parameters carrying phonological information. This spectrum includes, for example, Saito and
There is a PARCOR coefficient as described in "Rules for Audio Information Processing" by November II (published by Ohm Month). Note that this PARCOR coefficient is an example of a spectral parameter, and the present invention is not particularly limited to this.

On the other hand, the sound source analysis section 2 extracts sound source parameters such as fundamental frequency and amplitude information, which are sound source information, from the input digital audio signal. The sound source parameters extracted by the sound source 9111 section 2 are sent to the synthesis section 5 via the emotion circuit 4, and the spectral parameters analyzed by the spectrum analysis section 3 are also sent to the synthesis section 5. The synthesis unit 5 generates synthesized speech using the spectral parameters and the sound source parameters.

That is, the synthesizing unit 5 repeats the vector parameters in the intermediate frame at intervals of the reciprocal of the fundamental frequency (hitchi cycle) and converts them into a string of fame parameters. This synthesis section 5 includes, for example, J, D, Markel a.
nd A, H, GrayJrF5, using a lattice-shaped speech synthesis filter on the square paralysis 2 as described in Suzuki translation "Speech Linearity] Zdl" (published by Coronato 1),
A fame waveform is synthesized using the above-mentioned spectrum parameters and eight parameters. Note that the square multiplier lattice type voice dramatization filter is only one example, and other g/P synthesis means may be used. The synthesized iHf'j waveform is D
/A converter 6 and output as Issho-me 1.

Next, the emotion-adding lj operation of the above-mentioned vinegar pie i synthesis will be explained in more detail. Information 1[・1su9 circuit 4 is the sound source part F
Among the source parameters output from the ji section 2, the fundamental frequency 1 time t++1'* structure and the amplitude parameter are subjected to functional transformation, and the synthesis section 5 synthesizes a speech waveform using these sound source parameters and spectral parameters. Realized by

In other words, among the sound source parameters output from the source part + 11 section 2, ), (main frequency f. (1) is given λ j,
For example fo ' (t) -fo (t) -1-lr
(t), the new fundamental frequency f' (t
) to determine. Here, α(1) is a constant representing the increment of the fundamental frequency at each time. The functions listed here are examples and do not limit the invention. Also, regarding the time structure, the time series of minutes + 17 frames is
For example, Kitahara, Higashi Oko, “Temporal Structure and Stretching Characteristics of Speech” (Proceedings of the Acoustical Society of Japan (published March 1986))
) performs nonlinear compression depending on the rate of spectral change using LPC cepstral regression coefficients.

Of course, the method of expansion and contraction is not limited to this.

On the other hand, regarding the amplitude parameter P (L), a new amplitude parameter P' (t) is determined, for example, using the conversion formula P' (t)=P (t) x β (1). Here, β(1) is a constant representing the amplitude increment at each time. Of course, any conversion function can be used in this case as well.

The new sound source parameters converted as described above are sent to the synthesis section 5 together with the spectral parameters, and are outputted via the D/A conversion section 6 as a synthesized speech waveform.

FIG. 2 is a schematic block diagram showing another embodiment of the invention. In Fig. 2, an OC (not shown) is connected to the input terminal 7.
An input means such as R or a keyboard is connected, and the sentence −r;<ya! lj li! Something like "i" is input. To the input sentence
The morphological analysis unit 8 converts the `` and `` words into a string of morphemes based on the ``chu:go'' stored in the morphological dictionary unit 9 . The morphological dictionary section 9 stores at least the "pronunciation" and "part of speech", and for the input sentence string or symbol string, for example, "Kana-Kanji Conversion by Computer" by Aizawa and Jiang J +; Dividing into morphemes is performed using the ancient method of morpheme division, such as the technique (V + Study, 25-5). It is not limited to.

The 11 morphemes solved by +11 by the morphological analysis section r section 8 are given to the pitch control processing section 10, and
? Based on the contents stored in the old part 11 and the accent combination rule part]2, a fundamental frequency component representing the height of -8 is determined. The morpheme string with the fundamental frequency component attached is given to the gold coin parameter synthesis section 13, and is converted into a string of phoneme parameters based on the contents stored in the phoneme segment Y writing section 14.

The phoneme J-1 dictionary unit 14 holds phoneme fragments, that is, phonemes constituting bunka or words, or phoneme chains such as cv, vc, etc., as parameters, and phonemes constituting morphemes or cv, vc, etc. The phoneme segments are arranged in accordance with the phoneme order, and the parameters within the unit frame are repeatedly converted into a string of speech parameters at intervals of the above-mentioned pitch period. At this time, the fundamental frequency 1-temporal structure and amplitude parameters of the \I scale-declarative sentence are each subjected to, for example, the above-mentioned function transformation, and are output as new δ source parameters to the pitch control processing unit]0. Thereafter, the synthesizing section 16 and the D/A converting section 17 perform the same operation f'l as in the embodiment shown in FIG.

[Efficacy of the invention As mentioned above, the invention according to claim 1'I is:
It analyzes the input digital sound small signal to extract feature parameters, and expresses "anger" by controlling only the three elements of prosody, such as the fundamental frequency, temporal structure, and amplitude of the extracted feature parameters. As a result, it is possible to obtain a voice with natural emotion without significantly distorting the phonology, and it is also possible to control its intensity.

In the invention according to the second claim, the input character string or symbol string is converted into a pronunciation sequence, the converted pronunciation sequence is converted into synthesized speech, and the first i! ：'In the same way as the I term, the prosodic '
By controlling only the three elements, it is possible to produce speech with natural emotion without distorting the phonology, and it is also possible to control its intensity.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic block diagram of an embodiment of the present invention. FIG. 2 is a schematic block diagram of another embodiment of the invention. In the figure, ] is a digital audio signal input section, 2 is a sound source analysis section, 3 is a spectrum tli section, 4. ] 5 is 11 circuits with emotions, 5. ] 6 is a synthesis unit, 6.17 is a D/A conversion unit, 7 is an input terminal r, 8 is a morphological analysis unit, 9 is a morphological dictionary unit, 10 is a pitch system ■1 processing unit, 1-1 is an accent dictionary 12 is an accent combination rule section, ] 3 is a speech parameter synthesis section, and 14 is a phoneme dictionary section.

Claims (2)

[Claims] (1) An input means for inputting a digital audio signal; an analysis means for analyzing the digital audio signal input by the input means to extract feature parameters; and a synthesized speech using the feature parameters extracted by the analysis means. A speech synthesis device comprising a synthetic speech converting means for converting into an analog signal, and a D/A converting means for converting the synthetic speech converted by the synthetic speech converting means into an analog signal and outputting the synthesized speech extracted by the analyzing means. Synthetic speech emotion imparting characterized in that the fundamental frequency, time structure and amplitude parameters among the characteristic parameters obtained are functionally transformed, and a synthetic speech expressing the emotion of "anger" is generated by the synthetic speech converting means. circuit. (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 a pronunciation conversion means by the pronunciation conversion means. A speech synthesis device comprising a synthetic speech converting means for converting the converted pronunciation into synthetic speech, and a D/A converting means for converting the synthetic speech converted by the synthetic speech converting means into an analog signal, A synthesized speech emotion imparting circuit characterized in that a fundamental frequency, a time structure, and an amplitude parameter among characteristic parameters are functionally converted, and a synthesized voice expressing an emotion of "anger" is generated by the synthesized speech converting means.