JPS6331796B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a fricative sound recognition device for recognizing fricative sounds.

In recent years, research into voice recognition devices has become active, and some devices have been put into practical use and are now on the market. Also, expectations are increasing that it will become the most natural means of communication between computers and humans, which continue to advance rapidly.

However, a device that can correctly recognize the utterances of any speaker has not been obtained, and it is particularly difficult to detect and classify irregular consonants, and it is also extremely difficult to reliably detect and classify fricative consonants.

Furthermore, the conventional method requires a large number of arithmetic operations including multiplication and division, such as calculating a feature vector from the outputs of a large number of band filters and calculating frequency bias.

An object of the present invention is to provide a fricative recognition device that reliably detects and classifies fricative consonants of any speaker. Another object of the present invention is to provide a fricative sound recognition device that does not require multiplication and division in the fricative sound recognition unit and has simple arithmetic processing.

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a fricative sound recognition device according to an embodiment of the present invention. In the figure, reference numeral 1 denotes a voice wave detector, such as a low-noise close-talk type microphone. Reference numeral 2 denotes a laryngeal vibration detector, such as a vibration pickup, which is attached to the vocal cords of the larynx using medical double-sided tape or the like to detect laryngeal vibration. 3 is a palate contact detector that detects contact information between the tongue and the hard palate, and an example of its shape is shown in FIG. The palate contact detector 3 includes a plurality of electrodes 3
a are arranged in a row, and by attaching it to the hard palate in the oral cavity during use, it is possible to detect the state of contact between the tongue and the hard palate. FIGS. 3A to 3D are schematic patterns showing the state of contact between the tongue and the hard palate, and the hatched areas indicate the contact areas. Figure a is /
Patterns typically seen when uttering S, Z/,
Figure b shows a pattern seen when uttering /i, j, ∫/, etc.; figure c shows a pattern seen when saying /t, d, n/, etc.; figure d shows a typical pattern when saying /r/. This is a pattern seen in

Reference numeral 4 denotes a voice wave intensity detector, for example a detection smoothing circuit, which extracts the envelope of the voice wave. Reference numeral 5 denotes a high-frequency component detector of the audio wave, which is composed of, for example, a bypass filter circuit, an envelope extraction circuit such as the detection smoothing circuit in the intensity detector 4, and a division circuit, and extracts the envelope from the output of the high-pass filter circuit. By dividing by the circuit output, high frequency components that are not affected by audio intensity are extracted. Reference numeral 6 denotes a laryngeal vibration intensity detector, which is, for example, a detection smoothing circuit, and detects the envelope of laryngeal vibration. Reference numeral 7 denotes a temporary storage unit for detection information, which stores audio intensity information that is the output of the intensity detector 4, audio high frequency component information that is the output of the high frequency component detector 5, and output of the intensity detector 6. Laryngeal vibration intensity information and palate contact information, which is the output of the palate contact detector 3, are temporarily stored. Reference numeral 8 denotes a fricative sound recognition unit that recognizes fricative sounds based on information in the temporary storage unit, which will be explained in more detail below using FIG. 4.

In FIG. 4, reference numeral 81 denotes a sound section retrieval section, which searches for a section with a sound based on the voice intensity information A in the temporary storage section 7. Reference numeral 82 denotes a fricative sound testing unit that tests fricative sound candidates based on the voice high frequency component information in the temporary storage unit 7 when a voiced section is found. Reference numeral 83 denotes a first contact pattern inspection section, which tests the fricative candidate to determine whether it is a fricative or a non-fricative with a fairly high degree of accuracy based on the palate contact information stored in the temporary storage section 7. Reference numeral 84 denotes a voiced sound testing section, which tests whether the sound is a voiced fricative, a voiceless fricative, or a non-fricative based on the laryngeal vibration intensity information stored in the temporary storage section 7. 85 is a second contact pattern inspection unit, which detects /ts/ among voiceless fricatives based on palate contact information.
or /t∫/. 86
is a fricative section length testing section, which tests whether the fricative is one of the voiceless fricatives /S and ∫/ based on the high frequency component information B. Reference numeral 87 denotes a third contact pattern inspection unit, which inspects whether it is /S/ or /∫/ based on the palate contact information. 88 is a fourth contact pattern inspection section, which detects the voiced fricative /dZ/ or /dz/ or /Z/ based on the palate contact information.
Check whether /z/.

The phonemes uttered by the fricative recognition device configured as described above are /ts/, /t∫/, /S/, /
Which fricative is ∫//dZ/, dz/, /Z/, or /z/ is output.

Hereinafter, the operation of the fricative sound recognition section will be explained according to the flowchart shown in FIG.

(a) First, search for a voiced section. According to the sound intensity information in the temporary storage unit 7, if the sound wave intensity is larger than the threshold value determined through experiments and satisfies the conditions for duration length, for example, 80 msec or more, that section is determined to be a sound section. . If the above conditions are not met, it is recognized as silence.

(Sound interval search unit 81) (b) Inspect fricative candidates. Before and after the start of a voiced section, if the strength of the high frequency component satisfies the conditions for the threshold value and duration length determined through experiments based on the audio high frequency component information in the temporary storage section 7, that section is determined as a friction section. , the uttered phoneme is taken as a fricative candidate. If the above conditions are not satisfied, the sound is recognized as a non-fricative sound.

(Fricative sound inspection unit 82) (c) Next, a test is performed to see if it is a more reliable fricative sound candidate. Before and after the beginning of the friction section, there is a narrow contact pattern of SZ or IJ according to the palate contact information in the temporary storage section 7 (Fig. 3a,
b), and if the conditions for the duration determined through experiments are satisfied, the process proceeds to the next process as an almost certain fricative. If either the contact pattern or duration length is lacking, it is recognized as a non-fricative sound.

(Contact pattern inspection unit 83) (d) Inspect whether it is a voiced fricative or a voiceless fricative.
Based on the laryngeal vibration intensity information stored in the temporary storage unit 7, a section in which the laryngeal vibration intensity satisfies the conditions for the threshold value and duration length determined through experiments before and after the start of the voiced section is determined as a voiced section.

If the beginning of the voiced section precedes the beginning of the fricative section by a length of time determined experimentally, for example 20 msec or more, it is considered a voiced fricative. Conversely, if the start of the fricative section precedes the start of the voiced section by an experimentally determined time length, for example, 20 msec or more, it is considered a voiceless fricative. If it is neither a voiced fricative nor a voiceless fricative, it is recognized as a non-fricative.

(Voiced sound testing section 84) (e) Among voiceless fricatives, unvoiced affricates are tested. In front of the beginning of the friction zone, the palatal contact information is examined, and there is SZ or IJ, and behind it there is a closing contact pattern TDN or R (see Figure 3 C, d), and the continuation determined by experiment. If the condition regarding duration is satisfied, affricate/
Let it be either tS/ or /t∫/.

In the case of /tS/ or /t∫/, if there is an SZ contact pattern behind the closed pattern, it is recognized as /tS/, and if there is no SZ contact pattern, it is recognized as /t∫/.

If it is neither /tS/ or /t∫/, proceed to the next process.

(Contact pattern inspection unit 85) (f) If the friction section length is longer than the time length obtained by experiment, it is determined as a candidate for fricative sounds /S/ and /∫/, and the process proceeds to the next step. If it is short, it is recognized as a non-fricative.

(Friction section length inspection unit 86) (g) At the rear of the end of the friction section, SZ of the contact pattern is determined based on palate contact information (see Figure 3 a).
, and if the conditions for the duration determined through experiments are satisfied, it is recognized as a voiceless fricative /S/. If the above conditions are not met, it is recognized as a voiceless fricative /∫/.

(Contact pattern inspection unit 87) (h) Recognize voiced fricatives. Examine palate contact information anterior to the beginning of the friction zone and
Or, if there is an IJ, followed by a closing contact pattern TDN or R (see Figure 3 c, d), and the conditions for the duration determined experimentally are satisfied, the voiced fricative /dZ/ /
dn/, and in other cases /Z/ or /z/. Furthermore, if a contact pattern SZ exists behind the start of the friction section, it is set as /dz/ or /Z/, and in other cases, it is set as /dz/ or /z/, so that /dZ/, /
Recognizes dz/, /Z/, /z/.

(Contact pattern inspection unit 88) As described above, according to the present embodiment, a voiced interval is searched based on voice intensity information, fricative candidates are inspected based on voice high frequency component information and palate contact information, Voiced fricatives and voiceless fricatives are examined from voiced intervals based on laryngeal vibration intensity information and fricative intervals based on speech high frequency component information, and uttered phonemes are determined from palate contact information and friction interval length.
Fricative sound /tS/, /t∫/, /S/, /∫/, /
It is possible to recognize which one is dZ/, /dz/, /Z/, or /z/.

Note that in this embodiment, fricative candidates are inspected before being classified into voiced fricatives and voiceless fricatives, but fricative candidates may be inspected after being classified into voiced and unvoiced sounds. . Furthermore, the sound interval search section 81 and each of the inspection sections 82 to 88 may be configured to perform test classification without determining the conditions for each duration length, and finally determine the conditions for the duration length. can.

As described above, the present invention is configured to temporarily store voice intensity information, voice high frequency component information, laryngeal vibration intensity information, and palate contact information, and recognize fricatives based on this temporarily stored information.
It is possible to realize a fricative recognition device that can reliably detect and classify fricative consonants of any speaker, does not require multiplication and division in the fricative recognition unit, and has simple arithmetic processing.
This has a remarkable effect on speech recognition, etc.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a fricative sound recognition device according to an embodiment of the present invention, FIG. 2 is a diagram showing an example of the shape of a palate contact detector, and FIGS. 3 a to 3 d are diagrams showing contact between the palate contact detector and the tongue. FIG. 4 is a block diagram showing a specific example of the fricative sound recognition section, and FIG. 5 is a flowchart showing the operation of the fricative sound recognition section. 1...Audio wave detector, 2...Laryngeal vibration detector,
3...Palate contact detector, 4...Intensity detector, 5...
...High frequency component detector, 6...Intensity detector, 7
...Temporary memory section, 8...Fricative sound recognition section.

Claims (1)

[Claims] 1. means for detecting speech waves, means for obtaining speech intensity information and information on high frequency components of speech from the detected speech waves, and means for detecting laryngeal vibration;
means for obtaining laryngeal vibration intensity information from detected laryngeal vibrations; means for detecting contact information between the tongue and hard palate; It extracts fricative candidate candidates based on the temporary storage section that stores the temporary memory and voice high frequency component information, increases the accuracy of the fricative candidate based on palate contact information, and extracts voicing candidates based on voice high frequency component information and laryngeal vibration intensity information. A fricative sound recognition device comprising: a fricative sound recognition unit that separates and identifies fricative sounds after separating them into fricative sounds, voiceless fricative sounds, and non-fricative sounds.