JPS60194500A - Frictivity detector - 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 Industrial Application The present invention relates to a friction detection device for detecting the strength of friction from speech, and can be used for speech recognition or pronunciation correction for speech-impaired people.

Conventional Structure and Problems Conventionally, a device for detecting the strength of friction such as /8/ or /Z/ from audio is generally a microphone 2.-.

Frequency components characteristic of fricative sounds are often detected using a filter bank or arithmetic processing such as fast Fourier transform, but these methods require complex circuits and take a long processing time, making real-time performance difficult. There are some shortcomings.

In addition, conventional fricative consonant practice devices for people with speech disabilities detect the number of zero-crossings per unit time, focusing on the fact that fricative consonants have many high frequency components, and swing a meter according to the number of zero-crossings per unit time. There is something. This method has a simpler configuration compared to using the filter bank mentioned above, but the vowels /1/, /u/
etc., the formants higher than the second formant spread over a high frequency range, and the level is also high, so for example, when pronouncing [:/il, the boundary between the /f/ part and the /i/ part is clear. Therefore, it was difficult to accurately detect the friction property only at the /f/ portion.

OBJECTS OF THE INVENTION The present invention solves the above-mentioned conventional problems.1. Therefore, it is an object of the present invention to provide a fricativeness detection device that obtains an emphasized fricative output in a fricative consonant part in speech.

Structure of the Invention In general, fricative consonants are sounds produced by flowing exhaled air into a narrow space formed by the tongue and palate, generating turbulent flow, so their frequency ranges from 4 to 6.
It is a noise-like sound with a peak near kllz. On the other hand, vowels are the original sound generated by the vibration of the vocal cords.
The sound is generated by strengthening or weakening the frequency using a type of resonator made up of the nasal cavity, tongue, lips, etc.
It consists of harmonic components called formants, which are generated by the -] - resonance and characterize vowels. The general frequency spectral distributions of the vowel /i/ and the fricative consonant /f/ are shown in Figure 1a,
Shown in b.

The present invention utilizes the difference in the same wave number spectrum between vowels and fricative consonants, and uses a high-frequency component extraction circuit that extracts high-frequency components from speech and a low-frequency component extraction circuit that extracts low-frequency components from speech, so that the speech signal remains constant. A zero-crossing number detection circuit that detects the number of times the zero line crosses within time, a division circuit that divides the output of the high-frequency component extraction circuit by the output of the low-frequency component extraction circuit, and a circuit that divides the output of the division circuit and the zero This is a friction detection device equipped with a multiplication circuit that calculates the product with the output of the intersection number detection circuit, and divides the output of the high frequency component extraction circuit by the output of the low frequency component extraction circuit, and divides the value into the above zero. By applying this to the output of the crossing number detection circuit, it is possible to easily separate vowels and fricatives by emphasizing the difference between vowels and fricatives.

DESCRIPTION OF THE EMBODIMENTS The component extraction circuit 2 is a low frequency component extraction circuit, which is composed of a bandpass filter circuit and a rectification/integration circuit.

Reference numeral 3 denotes a zero-crossing number detection circuit, which detects how many times the audio wave crosses the zero line per unit time, and is composed of a comparator, a counter circuit, and a DA conversion circuit. 4 is a division circuit that divides the output of the high frequency component extraction circuit 1 by the output of the low frequency component extraction circuit 2, and 6 multiplies the output of the division circuit 4 and the output of the zero crossing number detection circuit 3. It is a multiplication circuit.

The operation of the frictional property detection device of this embodiment configured as described above will be explained below.

According to experiments, in order to effectively separate fricative consonants and vowels and detect fricativeness as shown in Figure 1, the bandpass filter characteristics of high frequency component extraction circuit 1 should be set to 4 kHz to 8 kHz low frequency component extraction circuit. 2 bandpass filter characteristics to 300
It is optimal to set the frequency to 11 Hz, and it is better to amplify the high frequency component by about 10 times in order to match the levels of the low frequency component and the high frequency component.

Figure 3 shows the operation of this embodiment when the voiceless fricative [/i] (a) is input.The voiceless fricative [/i] is the voiceless consonant /
It is a syllable consisting of f/ and the vowel /i/, but the vowel /i
The second formant of / is very close to the peak of the frequency spectrum of the consonant /f/, and /i/ is not affected by the preceding /f/. It is difficult to reliably separate /f/ and /i/ from the output. In contrast, the high-frequency components and low-frequency components of the voice
゛ - The ratio of the /f/ part is 10:1 or more, and the /i/ part has a ratio of 1 to 0.8:1, so the high frequency component extraction circuit 1 and the low frequency component extraction circuit When the output of 2 is divided by the division circuit 4, /f/ is 10 or more /i/ is 1 to 0.8, and by multiplying the result and the output of the zero crossing number detection circuit 3, the third As shown in Figure d, it is possible to detect frictional properties in which the difference between /f/ and /i/ is emphasized. The same is true for voiced fricatives [31] (foundation), but the ratio of high-frequency components to low-frequency components of the voice is /3/, which is 3 to 2:1, and /i/.

The ratio is 1 to 0.8:1. Therefore, when the outputs of the high frequency component extraction circuit 1 and the low frequency component extraction circuit 2 are divided, the output of the division circuit 4 will be 0.8 to 1 at the /end/ part and 2 to 3 degrees /i/. By multiplying this with the output of the zero crossing number detection circuit 3, a frictional output in which the boundary between /3/ and /1/ is emphasized is obtained.

Effects of the Invention The friction detection device of the present invention includes a high-frequency component extraction circuit for extracting high-frequency components from voice, a low-frequency component extraction circuit for extracting low-frequency components, a voice zero-crossing number detection circuit, and a circuit for detecting the number of zero crossings of voice. By providing a division circuit that divides by the area component and a multiplication circuit that calculates the product of the output of the division circuit and the output of the zero crossing number detection circuit, it is possible to obtain an emphasized fricative output in the fricative consonant part. Therefore, it can be easily distinguished from the vowel part, which has a great effect.

[Brief explanation of drawings]

Fig. 1 is a spectral distribution diagram of the vowel /i/ and the fricative consonant /f/, Fig. 2 is a block diagram showing the configuration of the fricativeness detection device in the second real sister example of the present invention, and Fig. 3 is the same real example. FIG. 3 is an operation waveform diagram for explaining the operation of the FIG. 1...High frequency component extraction circuit, 2...Low frequency component extraction circuit, 3...Zero crossing number detection circuit, 4.
...Divide circuit, 6...Multiply circuit. Patent applicant: Director of the Agency of Industrial Science and Technology Kawa 1) Hirobe Figure 1/l/

Claims (1)

[Claims] A high-frequency component extraction circuit that extracts high-frequency components from audio, a low-frequency component extraction circuit that extracts low-frequency components, and a zero-crossing number detection circuit that detects the number of times an audio signal crosses a zero line within a certain period of time. and a division circuit that divides the output of the high-frequency component extraction circuit by the output of the low-frequency component extraction circuit, and the output of the division circuit.
A frictional property detection device equipped with a multiplication circuit that calculates a product with the output of a recorded zero crossing number detection circuit.