JPS63234299A - Voice analysis/synthesization system - 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 [Industrial Application Field 1] The present invention relates to a speech analysis and synthesis method, and particularly to a speech analysis and synthesis method that allows the speech rate to be varied without changing the pitch of synthesized speech. [Conventional technology 1] Currently, speech analysis and synthesis devices are widely used for guidance services in stations and the like. It is also being considered for application to guidance for personal computers and word processors. In particular, when used for guidance, it is desirable that the speaking speed be variable according to the user's request. By the way, if you simply change the speaking speed by changing the playback speed, the pitch will also change, which is not only unnatural but also difficult to hear.To solve this problem, we have developed a method to change the speaking speed without changing the pitch. The conventional method of
It is disclosed in Japanese Unexamined Patent Publication No. 16295/1983. That is, this publication describes a technology that detects a pitch periodic waveform from audio data stored in memory and thins out the number of repetitions of the pitch periodic waveform at a certain rate, thereby making it possible to vary the speaking speed without changing the pitch. Disclosed. [Problem to be Solved by the Invention 1] However, in the conventional method described above, when the number of repetitions of the pitch periodic waveform is thinned out at a certain ratio regardless of whether it is a vowel or a consonant, the sound quality deteriorates due to the shortening of the consonant. The sound would be unclear to the listener, but no proper consideration was given to this issue. Accordingly, an object of the present invention is to provide a speech analysis and synthesis method that eliminates the drawbacks of the prior art and allows the speech rate to be varied without changing the sound quality. [Means for Solving the Problems] In order to achieve the above object, the present invention provides a speech analysis and synthesis method that changes the speaking rate without changing the pitch by thinning out the number of repetitions of the pitch periodic waveform. Measure the frequency of the waveform. The present invention is characterized in that it is determined whether or not to perform thinning processing on the audio waveform in accordance with the result of the measurement. In an embodiment of the present invention, the above measurement is performed by calculating the number of zero-crossing points of a waveform composed of a predetermined number of waveform samples. As another embodiment of the present invention, the thinning process is performed only when the frequency is lower than a predetermined frequency as a result of the measurement. As yet another embodiment of the present invention, the predetermined frequency is a threshold frequency that can distinguish between vowels and consonants, and the thinning process is performed only on the vowels. [Operation] The present invention focuses on the fact that vowels and consonants can be distinguished based on the frequency of the speech waveform, and measures the frequency of each speech waveform section stored in the memory, and uses the It is now possible to decide whether or not to perform thinning processing. In other words, if the frequency of the speech waveform is low, it can be determined that the speech waveform is a vowel section, and if the frequency is high, it can be determined that it is a consonant section. Therefore, the consonant section is not thinned out, but only the vowel section is thinned out. Make sure to execute
As a result, it is possible to prevent deterioration of sound quality during voice speed conversion by thinning processing, and to generate clear synthesized speech to the listener. [Example] Hereinafter, an example of the present invention will be described in detail with reference to the drawings. FIG. 3 shows a block diagram of a system for implementing the invention. 1 is an A/D converter that converts an analog audio signal input to the input terminal 5 into a digital signal. 2 is a memory that stores digital audio data. The digital audio data stored in this memory 2 may be the A/D conversion converter specific power itself, or may be the output obtained by compressing the waveform. 3 is a CPU that performs sequence control of the entire system. 4 is a D/A converter that converts the digital audio data (restored if compressed) read from the memory 2 into an analog audio signal and outputs it to the terminal 6. In this embodiment, the CPU 3 performs consonant discrimination based on one frequency measurement.
This is done by FIG. 1 shows that in this embodiment, the CPU
3 shows a flowchart of speech analysis and synthesis processing (speed conversion processing) executed by No. 3. First, in step 11, the frequency at a predetermined point, point a, in each voice section of the voice data stored in the memory 2 in FIG. 3 is measured. This measurement is the number of zero losses Za between point a and N samples before. This is done by calculating . Point a is N within one voice section.
The number of zero crosses may be calculated for each point by sequentially shifting point a backward. In the next step 13, it is determined whether point a belongs to the vowel interval or the consonant interval based on the result of the previous step. That is, the number of zero crosses Za obtained in the previous step 11
is compared with a threshold value zth for the number of zero crossings between a consonant and a vowel, and if Za<Zth, the section to which point a belongs is determined to be a vowel section. Conversely, if Za≧Zth, the section to which point a belongs is determined to be a consonant section. If it is determined in step 13 that it is a vowel period, step 1
The process proceeds to the thinning process in step 5, and if it is determined in step 13 that it is a consonant section, the speech rate conversion process ends without performing the thinning process. The thinning process in step 15 can be performed using conventional known techniques such as detecting pitch cycle repetition and controlling the number of pitch cycle repetitions. Next, the speech analysis and synthesis processing according to the present invention will be explained in detail by showing the speech waveform and zero cross points when pronouncing "siroi J" in FIG. The frequency at point a within the interval, i.e. a
Calculate the number of zero crosses Za from the point up to N points before the sample number. The value of this Za is 13, and the threshold value Zth is lO
If so, then Za≧Zth, and this section is determined to be a consonant section, and no thinning processing is performed. Next, the frequency at point b within the interval of speech “i”, that is, b
The number of zero crosses zb from the point to the point N samples before is calculated in the same way. In this case, the value of zb is 5, and the relationship Za<Z holds true, so the section to which point b belongs is determined to be a vowel section, and the thinning process is performed. In this way, for each voice section in Fig. 2,
When the process shown in Figure 1 is repeated, the consonant sound "r"
The thinning process will not be performed for the section ``, and the thinning process will be performed for the sections containing the vowel sounds "0" and "i". In the example, it is set to 10, but in reality it is determined by the relationship with the number of samples N.Also, in this example, processing is performed by the CPU, but it may be implemented in a general-purpose signal processing LS1, speech analysis and synthesis LSI, etc. [Effects of the Invention] As explained above, according to the present invention, the frequency of the audio waveform is measured and the thinning process is not performed for audio at a frequency where the thinning process causes sound quality deterioration. By doing this, it becomes possible to not only keep the pitch unchanged, but also to vary the speaking speed without causing deterioration in sound quality.

[Brief explanation of the drawing]

FIG. 1 is a flowchart of the speech analysis and synthesis process of the present invention, FIG. 2 is a speech waveform diagram for explaining the process of FIG. 1, and FIG. 3 is a block diagram of a system for implementing the present invention. It is.

Claims (1)

[Claims] 1. In a speech analysis and synthesis method that varies the speaking rate without changing the pitch by thinning out the number of repetitions of a periodic waveform of speech, the frequency of the speech waveform is measured, and the result of the measurement is A speech analysis and synthesis method, characterized in that it is determined whether or not to perform thinning processing on the speech waveform according to the speech waveform. 2. The speech analysis and synthesis method according to claim 1, wherein the measurement is performed by calculating the number of zero crossing points of a waveform composed of a predetermined number of waveform samples. 3. The voice analysis and synthesis method according to claim 1, wherein the thinning process is performed only when, as a result of the measurement, the frequency is lower than a predetermined frequency. 4. Speech analysis and synthesis according to claim 3, wherein the predetermined frequency is a threshold frequency that can distinguish between vowels and consonants, and the thinning process is performed only on the vowels. method.