JPS6078497A - Affricates sorter - 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 classifier that can be used for speech recognition or for correcting the pronunciation of speech-impaired people.

Conventional example tIIl) Problems with formation and combination Conventionally, plosive sounds are separated from the sound wave detected by a microphone, and the characteristic frequency components are processed using a filter bank or fast Fourier transform (FFT). However, there is no known method for reliably extracting each phoneme of a plosive as the duration of the plosive is as short as approximately 10 m5 to 20 m5.

OBJECTS OF THE INVENTION The purpose of the present invention is to improve the changes in the internal sound of two voices accompanying speech;
10 Air velocity and 1'' lip opening/closing are detected by each detector, and the phoneme /p/, /l/ of rupture 1''?

An object of the present invention is to provide a plosive classification device that uses the Ii:i grand harameter' characteristic of /ni/ as a standard parameter and compares it with the inactive parameter to identify and reliably classify plosives into each phoneme. .

Structure of the Invention The present invention provides an oral pressure detector and voice detection device that can directly detect the increase in oral pressure that occurs during speech by utilizing the relationship between the intraoral pressure, sound 1, mouth airflow, and the opening and closing of the lips. microphone, an airflow detector that detects the flow of breath associated with speech, and an upper lip.

A lip proximity detector that detects the proximity of the lower lip is used to extract all of the above-mentioned characteristic parameters associated with speech, and when plosives are uttered, the plosives are analyzed for each phoneme, /p/, /l/.

Since this device directly detects all the most basic parameters during speech, there are no false positives due to individual differences compared to conventional plosive detection methods, and reliable plosive classification is possible. be.

Description of examples Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a plosive classification device according to an embodiment of the present invention. In Figure 1, 1 is a microbon that detects the sound when speaking, and 2 is an oral pressure detector that detects changes in the atmospheric pressure in the oral cavity when speaking. It consists of vessels, etc.

Reference numeral 3 denotes an airflow detector 11 that detects the flow rate of breath accompanying speech, and can be constructed of, for example, a hot wire flowmeter or a condenser microphone. 4 is a lip proximity detector that detects the opening/closing state of the lips; for example, by attaching a magnet 411 to the upper lip and attaching a magnetic force sensing element 411 such as a Hall element 1'- to the lower lip, 1. The proximity of the upper lip can be detected. 5 is a detection circuit for detecting the rise of the voice from i (? from the voice output signal from the voice detection microphone 1) to the presence of the voice, which is composed of an integrating circuit and a threshold circuit; 6 is the output sound constant I1 of the voice rise detection circuit 5. 111 is a delay circuit that mixes the delay, 1 is the internal pressure detector 2, 11 air flow detector 3, and 1-1 lip proximity detector 4.
1) Store Shizu as a feature parameter for a certain period of time! This is a 11i baranotor storage circuit, which is composed of an A-D conversion circuit and a memory circuit.

8 is a standard parameter storage circuit that stores in advance the % signature parameters specific to each old convention of plosives (/I)/, /l/, /ni/) as standard parameters, and is a general memory circuit. Can be configured. Reference numeral 9 is triggered by the output of the delay circuit 6 which delays the signal from the audio rise circuit 5, and is triggered by the output of the delay circuit 6 which delays the signal from the audio rise circuit 5, and combines the characteristic parameters stored in the characteristic parameter storage circuit 7 and the plosive sounds stored in advance in the standard parameter storage circuit 7. This is a phoneme classification circuit that compares standard parameters of each phoneme and classifies each phoneme based on similarity, and is composed of a comparison circuit and the like.

FIG. 2 is a layout diagram showing the arrangement of signal detectors 1 to 4 in FIG. 1. In FIG. 2, the microphone 1 and the oral pressure detector 3 are placed in front of the mouth, and detect the flow velocity of the oral airflow associated with sound and speech, respectively. The intraoral pressure detector 2 is fixed to a hard tube in the 10 cavity with adhesive or the like and detects the atmospheric pressure in the oral cavity. The IN lip proximity detector 4 measures the distance between a person's lower lips by fixing a small magnet to the lower lip and detecting the full intensity of magnetic lines of force by fixing a magnetic force sensitive element such as a Hall element to the lower lip.

According to experiments, when speaking the plosive sounds /p/, /l/, /ni/, the internal pressure 9 airflow 1 lip position is different,
-K, and sounds other than plosives jRi (e.g. fB sounds and fricatives) are also 11°! li: Su,” 1st sf+I? Jt+
Characteristic parameters of 'i'(1' internal pressure, 11 QV,' +
The plosive sounds /p/, /l/, and /ni/ are possible depending on the position of the lips.

In other words, the general VC plosive sound completely closes the oral cavity or vocal tract with the tongue or lips and accumulates exhaled air, so the internal pressure is 1°-1.
When the 11' rises to a certain limit, an explosion occurs and an impulse-like sound and 11 qi b11; are generated. The location where the closure is formed differs depending on the rhyme of each plosive. For example, as shown in Figure 3a, when VC1/p/ is pronounced in full, the closure is created with /+1, and when / 1 / is pronounced in full, the closure is formed in the 3rd place. As shown in the figure, the tongue and front teeth KJ: make a closure. When pronouncing /to/, a closure is formed by the force at the back of the oral cavity as shown in Figure 3C. , phonemes other than plosives, such as fricatives, are produced by causing turbulence by pouring exhaled air into a narrow space (sebame) formed by VC and +-+ 2:4, as shown in Figure 3 (d). .

Therefore, with the onset of speech, a relatively gradual rise in the oral cavity 11] and the oral airflow are observed due to the gulp. When pronouncing vowels, as shown in Figure 2, the mouth is opened and no increase in intraoral pressure or impulse air flow is observed.

Figure 4 shows the states of the above three types of characteristic parameters at the time of each phonetic utterance in Figures 3a-d and Figure 2.
Store the 11δ point of 4. For example, when making a plosive /p/ utterance, as shown in Figure 3a, 1-1 close the lips to hold the exhaled air, then suddenly release the closure and utter the sound at the same time. As for the characteristic parameters before and after the rise of ``Tame-chopsticks''(1's``2''), the closing of the lips and the increase of ``1'' in the mouth L1] can be seen before the rise of the voice, as in Figure 4, and there is a sharp rise at the same time as the rise of the voice. With the generation of pulsating oral airflow,
[''A rapid decrease in internal pressure and opening of the lips are observed. In the case of the plosive /l/, as shown in Figure 3bIK-, the closure peculiar to the plosive is formed by the tip of the tongue and the upper front teeth, unlike the lips in /p/, so as seen in Figure 4B. The state of the characteristic parameters before and after the rise of the voice is the same as in the case of /p/, where the intraoral pressure rises by 1 and rapidly decreases, and a pulse-like airflow occurs, but in the case of /p/ Unlike 1], the lips are located far apart. Also, in the case of /to/, a
'53 Close with the tongue at the back of the oral cavity as shown in Figure clc.
i1; By being formed, j seen in Figure 4C
;Although there was a pulse-like airflow, /p
11. Before the start of the sound seen in / and /1/]. In -, y (and phenomena such as rapid descent and closure of the 1'' lips as seen in /p/ are not observed. Make a gap with the 1" difference and the tongue to make Vc smaller, and make a roux for the 4th figure.
At the same time as the voice rises, a relatively slow rise in intraoral pressure and the generation of oral airflow occur. For vowels, li-me is pronounced with large [ ], the fourth
As shown in Figure E, before and after the beginning of the sound, 1? Keru 1
No characteristic changes were observed in either the internal pressure (1 mouth airflow) or the groove position.

The present invention utilizes the phenomenon of 1' internal pressure detector 2, 1' air flow test + 1S8;:3, U lips proximity detector 4 to extract the signal as a feature parameter, and ” 4 from the double detection circuit 5. ! ' ``When the trigger signal that passes 'j' through the delay circuit 6 comes 1, it waits for a certain period of time 4'5.
' ? The parameters are stored in the parameter storage circuit 7. According to the experiment, Figure 4.

The inside of the mouth before the rise of the voice as seen in B y1-
, D lip closure varies depending on phoneme, but 100
Since the consonant part of Minosutomi (which lasts from m5 to 200 mS, and is a plosive sound after a vocalization) is from 2Q○ to 300 m5, the characteristic parameter 8 is circuit 7 ((is a plosive after a lily) It is sufficient to hold the booter for 500m5, and it is okay to delete old data one by one.When a voice is emitted, the voice is converted into an electrical signal by the microphone 1, and the voice rise detection circuit 5 converts it into an electrical signal. The transition from silence to fame is detected for a certain period of time (according to experiments, the fricative/f
/ After the longest delay (200m to 30oms), the phonological classification circuit 9V is powered as a tricycle signal. Tory)j
- When the signal is input, the phoneme part 11 circuit 9 stores the standard parameters in advance in the standard parameter storage circuit 8, and - By comparing the input feature parameters stored in the feature parameter storage circuit 7 with the sound ``2I'' point as a reference and calculating the similarity, the plosive sound from the sound 7 = middle is 7'r i"r. Classified as 1rii.

Effects of the Invention The present invention is applicable to both speech VC and lI (gurgling) 11, 1-1 changes in internal pressure.

The velocity of the oral airflow and the opening/closing status of the 0 and 11 lips were detected by each detector, and i'7. , jin-t) Seven 11.5 points 1
] 1 loss for each plosive after IJ (/I) / , /l
/ , / to /) K characteristic '1 detection H;
-C: L- When ki, n 1 no 1j is made by hand.'' From Kotori, who makes a complete comparison between the L standard rose make and the manually made characteristic rose make, it is more reliable ((Divides plosive i etc. into each phoneme and j'↓ Even if there is a plosive sound in the word, (il:k '11.+7 is always preceded by silence, so the transition from silence to ion ゛?'4) Click the point, (
The present invention, which performs all recognition as quasi-points, ensures that each plosive sound is
It can be classified into phonemes.

[Brief explanation of drawings]

Fig. 1 is a block diagram of one embodiment of the present invention (plosive sound classification 2g position), Fig. 2 is an embodiment of the state of each single detector used in the present invention (i, t &) Layout diagram showing I
Figures 3 & ~ d are typical phonological utterances + 15 articulatory styles. Figures 1 and 4 are diagrams showing the correspondence between each phoneme and the waveforms of each % mark and Z-lammeter during typical phonological utterances. It is. 1...Microphone 11 sound, 2...11 internal pressure detector, 3...[] Airflow detection: 4::, 4.
...1-1 Lip proximity 1&detector, 5...Voice rise detection circuit, 6...Delay circuit, 7...
...% characteristic parameter description 1, 6 circuit, 8...standard rose makeup memory circuit, 9...jjlj rhyme minute million'
"circuit. Patent issued 1. fffi person Itsuzuke Technical Director Kawa 1) Hirobe Figure 1 Figure 3 Figure 8 Figure 3

Claims (1)

[Claims] A voice detection microphone placed in front of the mouth, an internal pressure detector placed in the oral cavity, and an airflow detector 1 placed in front of the mouth to detect oral airflow associated with speech. , a lip proximity detector that detects the opening and closing of lips, a voice start detection circuit that detects a rise from silence to sound from the voice signal detected by the voice detection microphone, and a signal of the voice start lily detection circuit. a delay circuit that delays the
A standard parameter storage circuit that stores in advance characteristic parameters for storing characteristic parameters extracted by the air flow detector and the lip proximity detector for a certain period of time, and the output signal of the voice rise detection circuit in the I-1 internal detector 9 A plosive has a phoneme classification circuit which compares the contents of the characteristic parameter storage circuit with the contents of the standard parameter storage circuit for plosives using a signal delayed for a certain period of time as a trigger, and classifies each plosive into a phoneme. Part of power’! Device.