JPS61296396A - Voice code generation - Google Patents

Abstract

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

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field 1] The present invention relates to a speech code creation method for rule synthesis.

[Background technology 1] Conventionally, in this type of speech synthesis method, a character sequence is input, as well as information about the accent of a word and the intonation of a sentence. Performing synthesis.

However, with this conventional method, when inputting a sentence from the keyboard, it is necessary to input the accent position of each word at the same time, so it is extremely troublesome to fabricate sentences.

[Object of the Invention] The present invention has been made in view of the above-mentioned problems, and its purpose is to provide a method that makes it extremely easy to input accent information when creating a speech code for rule synthesis. is to provide.

[Disclosure of the Invention] The speech code creation method according to the present invention takes speech as input, converts the speech waveform into a character sequence using speech recognition technology, extracts pitch information of each phoneme, and converts the pitch information together with the character sequence. It is characterized by the ability to easily input character sequences and accent information by adding voice input to character input from a conventional keyboard, or by voice input alone. .

FIG. 1(a) shows an embodiment of the voice code creation method according to the present invention. In the figure, characters input from a keyboard or a character reader are broken down into phonemes such as phonemes and syllables and stored in i. Next, the speech input from a microphone or the like is segmented into phoneme units by the mouth, and at the same time, the resulting phoneme string is compared with the phoneme string from the character sequence, and if they do not match, it is segmented again. By redoing the process, the phoneme boundaries are accurately detected, and parameters such as the pitch, power, phoneme, and formant information of each phoneme are extracted in C, and these are added to the character information from the character sequence. Then, in step 2, encoding is performed.

The food obtained in this way is stored in a memory or printed as a barcode, and when compositing, the code is read out in E and decoded into each parameter in B, as shown in Figure (b). Then, in G, synthesis is performed using the phoneme data and rules previously stored in the synthesis section.

Therefore, in the above embodiment, phoneme segmentation can be performed easily and accurately by comparing phonemes obtained by speech recognition with known phoneme II sequences, and information regarding accents and intonations can be easily and accurately performed. It can be easily obtained from the input.

The embodiment shown in Figure 2 uses only voice input to segment the voice waveform into phonemes and convert them into character sequences.During this segmentation, pitch information and phoneme length are simultaneously analyzed. By extracting information such as, it is possible to omit the separate input of accent information from the keyboard, as in the case of FIG. Naturally, in this case, the accuracy of the speech recognition circuit is a problem, but recently, by using a pre-ambiguity processing method that the inventors have proposed separately, it has become possible to achieve relatively inexpensive and highly accurate speech. A recognition circuit can be configured.

[Effects of the Invention] As described above, the present invention creates a voice code for rule synthesis by character input and voice input, or by using voice input only, and converts voice waveforms into characters by segmentation. When converting to a series, pitch information is extracted at the same time, and this is used as accent information when composing rules, so it is possible to input characters and accent positions from the keyboard, which was conventionally done. There is an advantage that at least the input of the accent position can be omitted, and the creation of the voice code can be considerably simplified.

[Brief explanation of drawings]

1(a) and (b) are 70-charts showing one embodiment of the method of the present invention, and FIG. 2 is a 70-chart showing another embodiment.

Claims (3)

[Claims] (1) A speech code creation method which takes speech as input, converts the speech waveform into a character sequence using speech recognition technology, extracts pitch information of each phoneme, and encodes the pitch information together with the character sequence. (2) A character sequence having the same content as the voice input is input in advance, and pitch information is extracted while comparing the phonemes obtained by segmentation with the phonemes from the character input. How to create audio code as described. (3) Claim 1 characterized in that information regarding other prosodic or articulatory combinations is extracted along with pitch information.
How to create audio code as described in section.