JPS6078498A - Voiced sound section extractor - 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 is directed to an apparatus for extracting voice segments used for speech recognition, language training, etc.

Conventional fit) formation and its problems What is a voiced sound? A voiced sound is a sound emitted by the vibration of)4-H-'i;'. Or, if the voiced sound is correct ff(ii
Not only is it a basic elemental technology in speech recognition, but also measuring how the fit band vibrates in 1°4 rhymes is an important part of vocal training for people with disabilities. This is extremely important. Conventionally, voiced sounds have been extracted by removing harmonic signals and unvoiced π'' that affect the vocal tract characteristics using voice signal frequency analysis, etc. (referred to as pitch)
The common method is to extract the pitch and define the interval 17 where the pitch is extracted as a voiced section, but extracting the pinch accurately from the voice depends on individual differences, the influence of external noise, v1'4j
′? It is difficult to avoid the appearance of double pitches or missing pitches due to differences in style. Therefore, rather than extracting the vocal cord vibration from the voice picked up by a microphone, it is possible to
There are methods to extract the vibrations of the vocal cords through the outer wall of the trachea using an accelerometer or contact microphone, but this method is often more direct and effective. A block diagram of this example is shown in FIG.

In FIG. 1, numeral 1 denotes a laryngeal vibration detector, which is generally a sensor or contact microphone, such as an acceleration pickup, that extracts all vibration acceleration, and is fixed to the larynx with a nokundo or double-sided adhesive tape. 2 is a rectifier-integrator circuit that rectifies and integrates the signal from the laryngeal vibration detector 1 and converts it into a DC signal proportional to the energy of the signal. Reference numeral 3 denotes a threshold circuit J which cuts the value obtained by the rectifier/integrator circuit 2 using a constant threshold value Aq to extract a voiced section. Below is the configuration and problems of the conventional example. lr, 2
This will be explained based on the diagram. Waveform a in Fig. 2 is the output of the laryngeal vibration detector 1, but in addition to waveform a1, which is the vibration of the vocal cords detected through the outer wall of the trachea, there is also a waveform caused by the movements of the throat that occur when swallowing or moving the neck. a2 is also detected at the same time. The waveform S is the signal of waveform B converted into a DC signal by the rectifier/integrator circuit 2, and when the waveform S is cut at a certain threshold by the threshold circuit 3, the waveform C is produced. In this waveform C, 1. In addition to the original A 3-voice section due to the II + u movement of the j band, voiced output is also obtained in the throat movement VC, 1: - section VC, and the voiced section and the 11th 1 voiced section.
I often get 0 out. Waveforms caused by such movements of the throat (swallowing, full head movements) are not included in the audio signal shown in waveform d. do not have. Unlike speech, the force θ that extracts voiced sections directly from the throat is very effective because the movement of the vocal cords is more directly affected, but it also extracts signals caused by the movement of the throat. In addition, it often has the same frequency component as the vocal fold vibration, so it is difficult to know whether to separate it with a filter or the like, and to correct it with a real-quick j. OBJECT OF THE INVENTION The present invention solves the above-mentioned conventional problems, and eliminates the waveform caused by the movement of the throat that is unrelated to the vibration of the vocal cords.・Provide a complete voiced section extraction device that can extract all voiced sections.

([−+”l).

Structure of the Invention The present invention provides a laryngeal vibration detection means fixed to the larynx for detecting the vibration of the vocal cords through the outer wall of the trachea, and a laryngeal vibration detection means placed in the oral cavity to detect vibrations in the cavity during closure during articulation and the formation of a narrowing. 1-1 internal pressure detection means for measuring changes in atmospheric pressure; 1-1 internal pressure detection means; 1-1 internal pressure detection means for detecting an area in which the intraoral pressure is increasing based on the output of the internal pressure detection means; The output of the laryngeal vibration detection means is controlled by the voice section detection means which detects the section where the voice is present from the voice output of the detection means, and the output of this oral section detection means and the 8-nola section detection circuit. It is equipped with two gate means for passing the laryngeal vibration output only in the section where the voice output is present, an addition means for adding the outputs of the two gate means, and a voicing detection means for detecting the voiced section from the magnitude of the output of the addition means. This is a voiced section extraction device that extracts only the parts related to the laryngeal vibration waveform, vocal articulation, and actual speech output, and removes all the waveform parts caused by throat movements, etc.
This makes it possible to extract the voice section without error.

DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 3 is a diagram showing a voiced section extracting device according to a second embodiment of the present invention. In Fig. 3, 4 is the micro-bore that detects the utterance 11 and the 1'4 voice of te.
5 is the utterance + 1y voice, 11; vibration (7, fH: Laryngeal vibration calibration that is detected through the entire outer wall: (;;, and is composed of, for example, a pickup or a contact microphone. 6 is (1) is an internal pressure detector that detects changes in the air inside the cavity, and is made up of, for example, a small EE force detector such as a semiconductor strain cage.7 is from the microphone 4 for sound detection. ) °) Voice volume Chi 1-1 - Preservation of voice from bow (I J' - f', r 'lj
s 7i, j4), and is composed of a rectifying and integrating circuit and a threshold circuit.
The V unit 11 internal pressure detection circuit is a circuit for detecting an interval in which an increase of 1 in 11 from the output of the internal pressure detector 6 is recognized, and is constituted by a threshold circuit. Reference numeral 9 denotes a first gate circuit, which allows the output of the laryngeal vibration detector 6 to pass through only the section recognized as a sound section by the output of the voice section detection circuit 7, and can be easily configured with an analog switch circuit. .

Reference numeral 10 denotes a second gate circuit having the same configuration as the first gate circuit 9, and is a circuit that turns on the output of the laryngeal vibration detector 6 only in the section where the intraoral pressure rising section is recognized by the internal pressure detection circuit 8. . 11 is an adder circuit, and the first, ? -1-Circuit 9
and the output of the second gate circuit 10 are added to '=. Reference numeral 12 denotes a voiced detection circuit which extracts a voiced section from the output of the adder circuit 11, and is composed of a rectification and integration circuit and a threshold circuit, and when the rectified integral value exceeds a certain threshold value, it is determined that the voiced section is voiced.

FIG. 4 is a layout diagram showing the arrangement of the respective bow detectors 4, 6°6 in FIG. 3. In Figure 4, microphone 4
is placed in front of the mouth to detect sound. Oral pressure detector 6
The laryngeal vibration detector 6 is attached to the larynx with an adhesive such as double-sided adhesive tape or a band worn around the neck. Fix it with something and detect the vibration of the vocal cords.In Fig. 4, 13 is the tooth, 14th is the 4.1. cavity, 16 is the l-1 lid, 1
6 is the cavity, 17 is the tongue, and 18t is the lips.

The voiced section detecting device of the present embodiment F11 configured as described above will be described below.

According to the experiment, it is 415. ．． The relationship between laryngeal vibration, voice signal, and internal pressure waveform at 111 o'clock V differs depending on the phoneme, and for lU' sounds and unvoiced sounds, voice and laryngeal vibration rise almost simultaneously in time, and simultaneous VCr'l', 'I'−)'J”
Ru. As an example, FIG. 1 shows the waveforms of various parts in FIG. 3 when the IEJ sound lal is produced. The waveform [is the output of the throat νI′l vibration detection unit 5, and has a waveform A1 due to the movement of one barrel and a waveform −d2f due to the vibration of the vocal cords (17 skin type
j is the output of microphone 4), and the waveform is 1 in 1-1.
1- shows the output of detector 6; At this time, the gate circuit 9 is opened only in the section where the output of the voice section detection circuit 7 is on, and the laryngeal vibration waveform -16=j1+ is generated, so that the movement of the throat as seen at the beginning of the waveform A is generated. The waveform caused by V is removed from the output of the gate circuit 9 with waveform E.

10 Internal pressure is 1" J sound etc. 1-1 i open";j'5
In the phoneme with 'i:s', it hardly increases by 1, so there is no change in the waveform no. Therefore, as shown in the waveform 1-1, the output of the internal pressure detection circuit 8 is also off, and as shown in the waveform 1-1, the gate circuit 10 is closed and no output is output. Therefore, the output H of the adder circuit 11 becomes only the output H of the gate circuit 9, and the voicing detection circuit 12 removes the 1-head vibration waveform unrelated to voice generation such as throat movement, and detects the voiced section corresponding to the vocal fold vibration. can be detected.

On the other hand, the operation when uttering a voiced plosive, for example 1bal, will be explained with reference to FIG. i″L by experiment
The relationship between voice, laryngeal vibration, and intraoral pressure in voiced plosives is
First, the tip of the tongue 17 and the peak 13 in FIG.
Laryngeal vibration called Z precedes, but in 9, Bu
In many cases, ZZ is difficult to distinguish between minute gaps. The increase in 1-1 internal pressure due to closure suddenly decreases due to the release of exhaled air during rupture, and the following vowels are disused, but the sound and throat vibration of the vowel (Xb) The internal pressure is the same as in FIG. In the BuZZ section up to the burst, the level of the sound detected by the microphone 4 is often not sufficient, and the noise section detection circuit T K :J=・t'f
)”・The interval is 18. Therefore, the gate circuit 9 is closed as the waveform is small.
On the other hand, the L1 Mo in 11 was promoted by 1 in the BuZZ section.
As shown in the waveform 11, the output of the internal pressure detection circuit 8)
J is turned off, the gate circuit 10 is opened, and the waveform 1・V
As shown in C, laryngeal vibration in the BuZZ region was detected.11. Ru. In the 14th part of 1 Mugi Zoku n no, this relationship is reversed: 'Z
)-'・The output of section detection circuit 7 is on (waveform 2), and 1
[) The output of the old 1 detection circuit 8 is turned off (to the waveform), and laryngeal vibration is detected only in the waveform V (as shown, t, lT!'' section. In the addition circuit 11, the gate circuit 9
, 'By adding all the outputs of I O, all the waveforms due to throat movement, like waveform Q, are removed and only the vocal cord vibrations are left.'', /
2. The laryngeal vibration waveform can be extracted, and the voicing detection circuit 12 can extract the accurate i'A interval from this waveform.

, day(niji-+friction j′ truth), 1Zal pronunciation gold side 7@ and explain.; i name] Narai t, j
, rupture γ? To, 1. Toy, fortune-telling device: 4. '51＠° forms a gag and V ko ]) Since this is an articulation method that creates a turbulent flow by letting air flow in, the intraoral pressure is gradual throughout the fricative consonant section, as shown in the waveform C. Indicates rising.

In the fricative consonant section, the sound signal shown at the beginning of the waveform shows a waveform with a harmonic structure due to vocal fold vibration as well as high-frequency components unique to fricatives, but the level is small compared to the following vowel and the duration is rainbow-like. In many cases, the voice section detection circuit 7 does not recognize a voiced section, and therefore laryngeal vibration is detected only in the following vowel section from the speech signal, as seen in waveforms 2 and 2. On the other hand, as mentioned above, in the fricative consonant section, an increase in the internal pressure of 1" is observed, so the output of the oral pressure detection circuit 8 is turned on, and the laryngeal vibration in this section is output as shown in waveform 1. is added to the output of the subsequent vowel part (waveform E) in the adder circuit 11, and then added to the output of the following vowel part (waveform E).
Accurate voiced sections are extracted from which waveforms due to throat movements are removed.

As described above, according to this embodiment, the opening/closing of the gate circuit is controlled by the presence detection signal from the voice waveform detected by the microphone of laryngeal vibration and the detection signal for the period of increase in oral pressure from the oral pressure detector. (Yo1, throat movement VcJ, throat!7+'i l) Remove the linac waveform. From the mutter laryngeal oscillation, it is possible to extract the voiced section corresponding to the movement of the vocal cords.

Effects of the Invention The voiced interval extraction device of the present invention includes a laryngeal vibration detector fixed to the larynx to detect the vibration of J:1 through the outer wall of the trachea, and a laryngeal vibration detector placed inside the l l l to detect the closure during articulation. Formation of a glen K f'l' 11 Intraluminal Qi 1i:, To measure all changes in 1'' within 1) - Detection: (:; and the sound j!
- Detection microphone ['' pops, Li inside l-1 [''! comes out;, 1-'' internal pressure rises due to the appearance of the wound.

A circuit that detects the 1 section where the sound is on, a circuit that detects the 8' interval from the sound, and a circuit that opens the gate only in the 1 rising section and the A J section from the sound. Pass vibration 2
The laryngeal vibration waveform is , vocal key 1'? By extracting only the part related to the actual voice production and removing the waveform part caused by throat movement hfFK, it is possible to accurately extract all voiced sections in real time.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a conventional voiced section detection device using 111954j'i vibration detection, Fig. 2 is a waveform diagram showing the operation of the same device, and Fig. 3 t is a corner extraction device according to an embodiment of the present invention. Figure 4 shows Example 1 of 1'i.
Figure 6 is a diagram showing the arrangement of various detectors, and a waveform diagram for explaining the invention in detail.
FIG. 6 is a waveform diagram for explaining the present invention in detail when a voiced plosive is uttered, and FIG. 7 is a waveform diagram for explaining the present invention in detail when a voiced fricative is uttered. 4...Microphone, 6...Laryngeal vibration detector, 6...(-1 internal pressure)
Arj, 7... Voice section detection circuit, 8...
-1-1 internal pressure detection circuit, 9, 10...gate circuit, 11...addition circuit, 12...voiced detection circuit. 4,! Cod '1 Applicant L7-Director of the Technology Agency Kawa 1) Hirobe Figure 1 Figure 3 Figure 4

Claims (1)

[Claims] a voice detection means for detecting voice; a voice section detection means for detecting a section in which voice exists from the output of the voice detection means; a laryngeal vibration detection means fixed to the larynx for detecting vibration of the vocal cords; an intraoral pressure detecting means arranged to detect the intraoral pressure; an intraoral section detecting means detecting a section where the intraoral pressure is higher than a predetermined value from the output of the intraoral pressure detecting means; and the laryngeal vibration detecting means. a first gate means for controlling the output of the voice section detecting means according to the output of the voice section detecting means; a second gate means controlling the output of the laryngeal vibration detecting means according to the output of the oral section detecting means; and a voiced section extraction device, comprising an adding means for adding the outputs of the second gate means, and a voiced section detecting means for detecting a voiced section from the output of the adding means.