JPS59162597A - Voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a speech synthesis device for synthesizing speech waveforms from sentences.

Synthesizers have been put into practical use. As a result, the number of words required has increased, and furthermore, there has been a demand for the development of voice response systems that can synthesize arbitrary words. In such a voice response system, it is very advantageous in terms of storage capacity and transmission speed to store and transmit the response word or sentence as a character sequence.

Speech synthesis devices used in such voice response systems include a speech synthesis device that synthesizes word speech each time it receives a word (hereinafter referred to as a word speech synthesis device), and a speech synthesis device that takes a sentence as input and analyzes the entire sentence. A speech synthesis device (hereinafter referred to as a sentence speech synthesis device) that synthesizes speech by adding prosody as a sentence has been developed.

Hereinafter, a conventional speech synthesis device as described above will be explained with reference to the drawings. FIG. 1 shows the configuration of a conventional word speech synthesis device.

In FIG. 1, 1 is an input unit that receives words; 2 is a word analysis unit that analyzes words to generate spectral envelope information S, duration information, amplitude information A, and pitch information P; 3
2 is a synthesizer which generates a speech waveform by inputting the outputs S, D, A, and P, and 4 indicates an output of the synthesized speech waveform.

In the word speech synthesis device configured as described above, a word received by the input unit 1 is immediately analyzed by the word analysis unit 2, and the word is analyzed by the word analysis unit 2, and is converted into spectral envelope information S (for example, Percoll coefficient, time series of formant frequency and bandwidth). Duration information, amplitude information A, and pitch information P are generated. Then, by inputting the above S, p, Afz and P, the synthesis section 3
The speech waveforms are synthesized in , and a synthesized speech waveform output is obtained.

However, the word speech synthesis device configured as described above synthesizes word speech every time it receives a word, so when a sentence with consecutive words is input, it cannot give a natural pinch as a sentence. has.

FIG. 2 shows the configuration of a conventional sentence-to-speech synthesis device. In FIG. 2, 6 is an input unit that receives a sentence, 6 is a storage unit that stores the sentence input to the input unit 6, and 7 is a sentence unit that analyzes the sentence and generates a sentence fundamental frequency subtraction turn FB. An analysis unit 8 analyzes a word and generates spectrum envelope information S, duration information, amplitude information A and word fundamental frequency contour (turn FW); 9 a word analysis unit that generates the above FS and the above F;
When pitch information P is generated from W, a pitch information generation unit,
10 is a synthesis unit that generates a speech waveform by inputting the above-mentioned S, p, A, and P, and 11 is a synthesized speech waveform output.
This shows that.

In the sentence-to-speech synthesis device configured as described above, sentences received by the input section 6 are stored in the storage section 6. The sentence-folding unit 7 analyzes the sentence stored in the storage unit 6 and generates a sentence fundamental frequency contour (turn Fs) according to the type of sentence. The words are read and analyzed, and spectral envelope information S, duration information, amplitude information A, and word fundamental frequency contour Fw are generated for each word.Next, the pitch information generation unit 9 generates spectral envelope information S, duration information, amplitude information A, and word fundamental frequency contour Fw. Sentence fundamental frequency contour pattern FS and word analysis unit 8
Pitch information P to be supplied to the synthesis unit 10 is generated from the word fundamental frequency contour (F) and turn FW generated in . Furthermore, the synthesis unit 10 synthesizes a speech waveform using the spectrum envelope information S generated by the word analysis unit 8, the duration information, the amplitude information A, and the pitch information P generated by the pitch information generation unit 9. An audio waveform output 11 is obtained.

However, in the sentence-to-speech synthesis device configured as described above, synthesis cannot be started until the entire sentence has been input. Furthermore, it takes a lot of processing time to analyze the entire sentence. As a result of the above, there is a drawback that a large amount of waiting time occurs from the start of inputting a sentence until the output of a synthesized speech waveform is obtained.

Purpose of the Invention In view of the above-mentioned drawbacks, the present invention provides a speech synthesis device that synthesizes speech with a natural pitch as a sentence, and that has a constant waiting time from the start of inputting a sentence to obtaining a synthesized speech waveform output. It provides:

Structure of the Invention The present invention comprises an input section that receives sentence control information and a sentence, a discrimination section that discriminates between the sentence control information and the sentence, a word analysis section that analyzes words included in the sentence, and a pitch information generation section. and a speech synthesizer equipped with a synthesis section that generates a speech waveform,
It is possible to compose speech with a natural pitch as a sentence with very little waiting time.

DESCRIPTION OF EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 3 shows the configuration of a speech synthesizer according to an embodiment of the present invention. In FIG. 3, numeral 12 denotes sentence control information consisting of information indicating the type of sentence (hereinafter referred to as sentence identification information) and information controlling each phrase included in the sentence (hereinafter referred to as phrase control information). 13 is a determination unit that determines the sentence control information input to the input unit 12 and the sentence;
14 is a word analysis unit that analyzes words to generate spectral envelope information S, duration information, vibration IJ information A, and word fundamental frequency contour pattern FW; 15 is a pitch information generation unit; and 15a is phrase fundamental frequency contour pattern FP. 16b is a sentence fundamental frequency contour pattern generation section that generates a sentence fundamental frequency contour pattern FB; 15c is a pitch information calculation section that generates pitch information P from the above-mentioned FW, FP, and Fs; 16 is the above S, D and A
and a synthesizing section 17 for synthesizing a speech waveform from P and P. Reference numeral 17 indicates the output of the synthesized speech waveform.

The operation of the speech synthesis device configured as described above will be explained below.

FIG. 4 shows an example of input to the speech synthesizer of the present invention. In FIG. 4, 1 indicates sentence identification information, and 2 to 7 indicate phrase control information.

FIG. 6 shows symbols used for sentence identification information, and FIG. 6 shows symbols used for phrase control information, their meanings, and generated phrase fundamental frequency contour patterns.

Further, FIG. 7 shows an outline of the sentence fundamental frequency contour pattern corresponding to the sentence identification information shown in FIG.

First, the input section 12 receives "0" and sends it to the determination section 13. Next, the determining section 13 determines the rOJ sent by the input section 12 as identification information indicating 1,000 sentences, and sends the sentence identification information Is to the sentence fundamental frequency translocation pattern generating section 16b. Next, the sentence fundamental frequency transposition number turn generation section 15b generates a sentence fundamental frequency transposition number turn of the plain text and sends it as an FB to the pitch information calculation section 15c. Next, the input unit 12 receives “6” and sends it to the determination unit 13. The determining section 13 determines "6" as phrase control information indicating the beginning of a phrase, and sends the phrase control information F to the phrase fundamental frequency contour pattern generating section 16a. Next, the phrase fundamental frequency contour pattern generation section 15a generates a phrase fundamental frequency contour pattern at the beginning of the phrase and sends it as an FP to the pinch information calculation section 15C.

Next, the input section 12 receives "6" and sends it to the determination section 13 for output. The determining section 13 determines "5" as phrase control information indicating the end of the phrase, and sends the phrase control information IF to the phrase fundamental frequency contour pattern generating section j5a. The phrase fundamental frequency contour pattern generation section 15a generates a phrase fundamental frequency contour pattern at the end of a phrase and sends it as an FP to the pitch information calculation section 15c. Next, the input unit 12 receives Th1sJ and sends it to the discriminating unit 13. The discriminating unit 13 discriminates [Th1s J as a word,
It is sent as is to the word analysis section 14. Next, the word analysis section 14 analyzes "Th1eJ, generates a word fundamental frequency contour pattern FW, and sends it to the pitch information calculation section 15c. At the same time, the word analysis section 14 analyzes the spectral envelope information S, duration information, and The amplitude information A is sent to the generation unit 16.The pinch information calculation unit 15c generates the word fundamental frequency contour pattern FW, the phrase fundamental frequency transposition pattern Fp, and the sentence fundamental frequency contour pattern yF3 as shown in FIG. The synthesizer 1 synthesizes it as pitch information P.
6.

The synthesis section 1e takes the supplied S, and! :A and P 1
-Synthesize the audio waveform corresponding to ThisJ to obtain a synthesized audio waveform output at 17. Similarly, ri, J and r-
Speech waveforms corresponding to J and "pen" are obtained in sequence.

As described above, according to the present embodiment, the discrimination section 13 that discriminates between sentence control information and sentences supplied from the outside, the phrase fundamental frequency contour pattern generation section 16a that generates a phrase fundamental frequency contour pattern, and the sentence fundamental frequency contour By providing the pitch information generating section 15 consisting of a sentence fundamental frequency contour pattern generating section 16b that generates a pattern and a pitch/pitch information calculating section 16c that calculates pinch information, by the time input of a sentence is completed,
It is possible to start synthesizing a speech waveform that has been given a natural pitch as a sentence.

In this embodiment, the phrase control information is placed immediately before the word that starts the phrase and immediately before the word that ends the phrase for each phrase of the input sentence, but the position thereof is not limited. For example, phrase control information may be placed immediately after a word that starts a phrase and immediately after a word that ends a phrase.

Further, the phrase control information of each phrase and the number of words included in the phrase may be placed immediately after the sentence identification information.

Furthermore, the number of words may be used to indicate the positions of the words that start a phrase and the words that end a phrase.

Effects of the Invention As described above, the present invention provides a discriminator that discriminates between sentence control information and a sentence supplied from the outside, and a pitch information generator that generates pitch information from the sentence control information. Synthesis of a speech waveform with a natural pinch added to it as a sentence can be started without waiting for it to finish, and this has great practical effects.

[Brief explanation of drawings]

1 and 2 are block diagrams of a conventional speech synthesis device, FIG. 3 is a block diagram of a speech synthesis device in an embodiment of the present invention, and FIG. 4 is a diagram showing an example of input in the same embodiment, FIG. 6 is a diagram showing symbols of sentence identification information and their meanings in the same embodiment, FIG. 6 is a diagram showing symbols of phrase control information, their meanings, and generated phrase fundamental frequency contour patterns, and FIG.
The figure shows an object fundamental frequency contour pattern generated corresponding to the symbol of sentence identification information, and FIG. 8 is a diagram showing calculations by the pitch information calculation section in one embodiment of the present invention. 12...Input section, 13...Discrimination section, 1
4...Word analysis unit, 16...Pitch information generation unit, 15a...Phrase fundamental frequency contour pattern generation unit, 16b...Sentence fundamental frequency contour pattern Generation unit, 16C...Pitch information calculation unit, 16.
...Synthesis department. Name of agent: Patent attorney Toshio Nakao and 1 other person
4 Figure 5 Figure 116 Figure 7 Figure 8

Claims (4)

[Claims] (1) An input unit that receives the sentence control information and the sentence, and a discriminator that receives the output of the input unit and discriminates between the sentence control information and the sentence, and analyzes the words included in the sentence to obtain spectral envelope information. , a word analysis section that generates duration information, amplitude information, and a word fundamental frequency contour pattern; a pitch information generation section that generates pitch information by receiving the sentence control information and the word fundamental frequency contour pattern as input; and the word analysis section. A speech synthesis device comprising a synthesis section that receives an output of the section and an output of the pitch information generation section and generates a speech waveform. (2) The speech synthesis device according to claim 1, wherein a sentence is represented by phonetic symbols or equivalent symbols and accent information. (3) The speech synthesis device according to claim 1, wherein the sentence control information comprises sentence identification information and phrase control information. (4) The speech synthesis device according to claim 1, wherein the sentence control information comprises sentence identification information and phrase control information.