JPS6031196A - Voice pattern generator - 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 The present invention relates to a voice pattern creation device, and more particularly to a device for creating voice characteristic patterns in a voice recognition device.

Figure 1 is a diagram for explaining an example of a conventional speech recognition device.In the figure, 1 is a microphone, 2 is a filter group, 3 is a sampling circuit, 4 is a speech section detection circuit, and 5 is a diagram for explaining an example of a conventional speech recognition device. The register 6 is a recognition unit, and as is well known, the voice input from the microphone 1° is frequency-analyzed by a group of band-pass filters 2, sampled at a constant cycle, and then distinguished between voice and noise to determine the voice interval. only is recorded in register 5. Alternatively, if the number of samples is large, the data is compressed and recorded in the register 5, and the recognition unit 6 uses the data. However, it has the disadvantage that it is difficult to distinguish between words that have different meanings depending on the sound at the beginning of the word, such as 6/roku/ and 1010ku/. In order to solve these drawbacks, a method has been proposed in which a detection circuit is provided to detect that the voice is in the rising state, and the sampling period of the sampling circuit 3 is thereby changed (Japanese Unexamined Patent Publication No. 1986-57). 191000), this method takes advantage of the fact that consonants are often located at the rising edge of speech, and samples rapidly changing consonants at a constant period shorter than vowels.
The aim is to fully extract the characteristics of consonants. Second
The figure shows the number sound 42/5htjyun by the above method.
This shows the audio signal and its sampling period when uttering "i/".As shown in the figure, the sampling period is 11 when the voice is starting, and the sampling period is 12 at other times, and t+ < t2. However, in reality, the energy of the voice is often not like this, and the example 6 and 10 million is shown in Figures 3 (a) and (b). In this way, the energy of a consonant is smaller than that of a vowel, and when a vowel is uttered, the energy rises. Also, the energy of consonants often decreases in the latter half, especially /r5k 1+ /, /o k
The /u/ of u/ is usually not uttered. In such a case, only the first half of /r/ and /ni/, where the energy is in a rising state, is sampled at a short period, and the second half is sampled at a long period. In addition, the final parts of vowels with long durations are also sampled in detail, so
It is not possible to fully extract the characteristics of consonants.

- Kazuma The present invention was made in order to eliminate the drawbacks of the prior art as described above, and in particular, it was made for the purpose of providing a device that creates a speech pattern by emphasizing the characteristics of consonants. It is.

Jujube - One Claw The structure of the present invention will be described below based on Examples.

The present invention focuses on the fact that voiceless consonants are made up of higher frequency components than vowels, and the sampling period is made finer when the high frequency components of the voice are larger than the low frequency components.

FIG. 4 is an electrical block diagram for explaining one embodiment of the present invention. In the figure, 7 is a comparator, and other parts having the same functions as those in FIG. The same reference numbers as in the case are given. In the speech recognition device shown in FIG. 4, the voice input from the microphone 1 passes through the band-pass filter group 2, but in the present invention, the highest band filter of the band-pass filter group 2 and The output of the lowest band filter is compared with the comparator 7, and when the output of the highest band filter is large, the sampling period of the sampling circuit 3 which samples the output of the filter group is shortened, and then only the section from the signal to the voice is sampled. I take it out and put it in register 5.

In the above embodiment, the highest and lowest band filter outputs are compared, but it is sufficient if the outputs are near the highest and lowest. It goes without saying that it is easy to understand that the same effect can be obtained even if the order of the parts in FIG. 4 is changed.

Shyness - As is clear from the above description, according to the present invention, it is possible to create voice parameters that weight consonant parts in words, which have traditionally been difficult to emphasize.

As a result, the recognition rate of the speech recognition device can be improved.

[Brief explanation of drawings]

FIG. 1 is a main part configuration diagram for explaining an example of a conventional voice pattern creation device, and FIGS. 2 and 3 are signal waveform diagrams for explaining the operation of the conventional device shown in FIG. , 4th
The figure is a configuration diagram of a main part 4 for explaining an embodiment of the voice pattern creation device according to the present invention. DESCRIPTION OF SYMBOLS 1... Microphone, 2... Filter group, 3... Sampling circuit, 4... Voice section detection circuit, 5... Register, 6... Speech recognition unit, 7... Comparator .

Claims (1)

[Claims] It has a means for frequency analyzing an audio signal, a means for sampling the frequency-analyzed signal, and a means for comparing the magnitude of the frequency components, and when the high frequency component is equal or larger in magnitude than the low frequency component, A voice pattern creation device characterized in that the sampling period of the sampling means is shortened.