JPS60257500A - Voice synthesization - 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 method for generating speech such as sentences and phrases by combining speech synthesis parameters in syllable units that have been filed in advance.

Structure of conventional examples and their problems Conventional speech synthesis methods of this type include a restricted synthesis method and a compromise method that uses only prosody extracted from natural speech.゛Regulated synthesis involves connecting pre-filed audio parameters for each monosyllable based on the rules for determining the duration of a clause in a sentence, and adding pitch information determined from the number of tones, etc. Create synthesis parameters using
This is used to synthesize speech. However, there are problems with the ease of listening and naturalness of synthesized speech because it is not possible to assign pitch information and the duration of syllables in a sentence according to rules. For this reason, compromise methods have been attempted, such as using syllable duration and pitch information in sentences extracted from natural speech. However, none of these methods incorporates the idea of controlling the rhythm unique to Japanese, so there is a limit to the ability to improve the audibility and naturalness of synthesized speech.

OBJECTS OF THE INVENTION The present invention eliminates the drawbacks of the above-mentioned conventional examples, and aims to synthesize speech such as phrases and sentences having a natural sense of rhythm by combining speech synthesis parameters in units of syllables.

Structure of the Invention In order to achieve the above object, the present invention performs syllable combination so that the listening timing point of a one-syllable file coincides with the listening timing point position extracted from natural speech. The effect is that a natural rhythm can be imparted to speech such as sentence spaces and phrases generated by combining the voice synthesis parameters of 'l-' order.

DESCRIPTION OF EMBODIMENTS An embodiment of the present invention will be described below. Figure 1 is an example of extracting the listening timing points for each syllable of the popular forecast (natural voice) [Kumori no Chiamede -1]. The procedure for extracting the natural speech listening comprehension point is as shown in FIG. Using this method, the listening timing point of natural speech can be easily and accurately extracted. This procedure will be explained below.

(1) A/J the natural voice from which the listening timing point should be extracted.
) Convert the computer's audio/no 1. into the file.

(2) Data in the audio file: FR, Extrusion, CI
t, T display, etc. to roughly cut out syllable sections from natural speech.

(3) Divide roughly determined syllable intervals in natural speech into analysis frames of, for example, 10m5 and perform LPO analysis (4)
In advance, the parameters of the listening timing point are read out from among the speech parameters for each syllable stored in the syllable parameter file, and the LP is calculated for each analysis frame.
Compare the degree of similarity with the voice parameters obtained by C analysis.

(5) Set the central point of the analysis frame where the degree of similarity is maximum as the listening timing point.

Figure 3 shows the natural γ? of Figure 3 determined by the above method. These are a speech synthesis waveform obtained by combining speech synthesis parameters by matching the listening timing point of each syllable with the listening timing point of the voice, and its listening timing (indicated by a square). Fourth
The figure shows a speech synthesis waveform obtained by conventional syllable combination and listening timing points extracted from it (indicated by squares).

A comparison between Figures 1 and 4 shows that the position of the listening timing point of the synthesized voice in Figure 4 is significantly different from that of the natural voice, and the synthesized voice of Figure 3 according to the present invention is more natural. You can see that you can get a sense of rhythm.

Effects of the Invention As described above, the present invention performs 1″? clause combination so that the listening timing point of the syllable file matches the listening timing point position extracted from natural speech.
It has the advantage of giving a natural rhythm to the sounds of sentences, phrases, etc. that are generated by combining 1'6-voice synthesis variations in syllable units.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the listening timing points for each syllable of natural speech, Fig. 2 is a diagram showing the listening timing point extraction procedure of natural speech, and Fig. 3 is a diagram showing the listening timing point extraction procedure for natural speech. A diagram showing a synthesized speech waveform. FIG. 4 is a diagram showing a synthesized speech 7h waveform obtained by a conventional method. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
Figure [River δ]

Claims (1)

[Claims] A speech synthesis method characterized by performing syllable combination by matching a listening timing point of a syllable file to a listening timing point position extracted from natural speech.