JPS59197100A - Voice wave 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 The present invention relates to an audio wave detection device for detecting human speech.

With the development of computers, voice input devices have been proposed. This voice input device stores the voice waveform of a specific person input into the computer in advance in the voice storage section, and recognizes the input voice by comparing the voice waveform uttered by the specific person with the value stored in the voice storage section. be.

This voice input device has various problems in that if a specific person catches a cold and the voice waveform differs from the memorized voice waveform, the voice cannot be input even if the correct voice is input.

In view of the above points, it is an object of the present invention to provide a sound wave detection device that detects sounds coming from the nose and mouth separately, thereby making it possible to obtain more accurate sound input. It is something.

That is, the present invention provides two audio reproducers at a predetermined interval at the tip of an arm fixed to the head, and one audio reproducer detects nasal output and the other detects oral cavity output. This is a characteristic feature.

The present invention will be described below based on embodiments shown in the accompanying drawings.

As shown in FIG. 1, a support arm 4 that supports two microphones 3m and 3n at its tip extends from a head animation 2 fixed to a person's head 1 in front of the face in an adjustable position. . The microphone 3n faces the nose to detect nasal output, and the microphone 3m faces the mouth to detect oral cavity output.

Note that it is desirable that either one of the microphones 3m and 3n is provided so that its position can be adjusted. Also, microphone 3m and 3m
It is desirable to provide a shielding plate 5 between the nasal passage and the oral cavity output to separate the nasal cavity output and the oral cavity output.

The signals from the microphones 3m and 3n are input to an AD converter 7 via a switch 6 such as a multiplexer, as shown in FIG.
performs processing for voice recognition.

The present invention has the above configuration, and an example of speech recognition will be described in detail below.

Figure 3 shows the energy curves obtained by microphones 3m and 3n, and if the energy curves from microphones 3m and 3n are NEm + N''n, /
a/ , /ka/ , /Sa/, /El/ , /n
a/, /ha/, /ma/, /ya/, /ra/, /
Energy curve N”m + of Wa/, /pa/, /N/
NEm is shown in Figures 3(a) to (I).This energy curve NEm + NEn is
, 3n is a time change curve of energy normalized by the maximum value.

Next, the characteristics of the energy curve shown in FIG. 3 will be described.

At /L/ and /h a/, a peak due to burst airflow appears when NEn and NEm rise at the same time, and IN'm is NE
It is rising faster than n. In /s a/, a small value (arrow) appears in NEm in the /s/ interval.

At /n/ and /ma/, NEn rises earlier than NEm, and NEn begins to decrease at the same time as NEm begins to increase. For /ya/, /ra/, /wa/, NEn and NEm rise almost simultaneously, but the slope at the rise is NEn.
is larger than NEm. In /N/, the oral cavity output is extremely small, and in NEm an energy curve of indoor noise appears.

The characteristics of the energy curve of each phoneme have been described above, and the parameters representing these characteristics are defined by the following equation tl), (2+) with delay time and slope ratio S.

D""'nO-'mo (1
) However, 'no l'mo is the time when NEn and NEm each exceeded the 5th point for the first time, (m3 is the time after, for example, 19.2m5ec from tlnO.Equation (1)
is the rise time difference between N”n and NEm, and equation (2) is NEm
It represents the ratio of the slopes of NEn and NEm at the rise of .

Figure 4 shows 10 types of monosyllables from Figure 3, excluding /Sa/ and //N/, plotted on the D-5 plane defined by these two parameters. I used what I said out loud. /Sa/teha/S/ NEm
It was excluded because the variation in the value of 'mo was wide, making the detection of 'mo unstable, and because the oral output was extremely small in /N/, mo could not be determined. According to Figure 4/
The delay time of -&/ and /h a/ is small, and the slope ratio S is distributed around 1.0.

For /ka/, /la/, /pa/, D) Q and S are small.

Also, for /na/, 7m a/, D<O and S<0, and for /y a/, /ra/, /wa/, D
is small, and S is larger than other phoneme groups. In Figure 4, S
) 3.0, /ya/, /ra1, /Wa/ plotted at the position of S = 3.0 are voice samples where NEm is slightly later than the rising edge of NEn (D<O'). It is shown in Figure 4 that there is. In this case, N
The outlines of En and NEm are the same as in Figure 1, but /ya/ in Figure 1.

も (arrow in Figure 4).

As shown in FIG. 1, the energy passing through the pharyngeal cavity 9 is Eo
([)l The output energy of the nasal cavity 10 and the oral cavity 11 is E
+1 (') + % (') Also, the energy observed by microphones 3n and 3m is En (in) (L), Em (i
n) (t), consider the problem of estimating the En (')/Eo (t) time-varying curve from the observed value En (in) (L). Assuming that energy is lossless within the vocal tract, equation (3) holds true.

I! ,o(') −En(t)''%(')
(3) Also, if Cn and Cm are the ratio of the radiated energy that enters each microphone, then CnEo (L) = En (Hn) (t) + (Cn/Cm
) Em(in)(') (51 is obtained. Also, /
The 10 utterances of ma/ were observed with one microphone placed at one end of the same cylinder covering the nose and mouth, and the normalized energy was NEol[)i (i = 1 to 10), their average. Assuming that the curve is NE 01c), 2 is calculated using the following equation. This result is expressed in a graph as shown in FIG. 5, and since the above value cn/cm can be determined from this, it can be obtained as En(t)/Eo(t').

Here, if we define the following equation as a parameter representing the characteristics of the En/Eo curve, (En/'o) o -En (to)/Eo(to)
16) △(En/Eo)-Max(En
(')/Eotl-En(”)/Eo(”)
](7) becomes. However, Max means the maximum value within the square brackets. Also, 0 is E. The time when t exceeds the 15% point of the maximum value for the first time, t' is, for example, 19.201 I
This is the time after f e C. Equation (6) represents the left end value of the En/Eo curve, and Equation (7) represents the maximum slope of the curve. Figure 6 is a plot of 12 types of monosyllables uttered in isolation by 10 men (the same speech samples as in Figure 4) on the plane defined by these two parameters (this is called the 0-△ plane). . (In order to make the diagram easier to read in Figure 5, there are phonemes plotted for only five speech samples, but the other five have a similar distribution.) In Figure 6, the above-mentioned phonemes are plotted on each speaker's side. method, we determined Cn/Cm and calculated the formulas +61 and +7+.As a result, 12 types of monosyllables (/a/, /ka/, /sa/, /la/, /
ha/, /pa/), (/na/, /ma/),
(/ya/, /ra/, /wa/).

(Classified into 4 groups: /N/'I).

Furthermore, Figures 7 and 8 show the distribution of parameters when appropriate correction is made to tno, and this correction is
As shown in Figure 9, 35 of the energy curve NEn
Connect the % point and 15% with a straight line, and set the intersection with the time axis as tno.
That is.

Next, as a parameter representing the characteristics of the waveforms of fricatives and affricates, we use the noise removal difference zero crossing number
ected differential zero c
The analysis using the rossing rate will be explained below.

When the oral cavity output audio signal is (Xl), the noise removal difference zero crossing number is defined by the following equation (11).

-1) Here, for each monosyllable, the time difference between the rise time of the noise removal difference zero crossing number and the rise time of NEn (note this, the 10th
As shown in the figure, the number of noise removal differential zero crossings (N, R-D, Z
, C, R) on the vertical axis and the rise time difference (Delaytime) between the noise removal differential zero crossing number and NEn shown in the equation
By plotting each single syllable according to its subsequent consonant (N-D plane), which has G as its horizontal axis, it is possible to distinguish between fricatives and affricates.

NEQ I NEm rise time 'ilQ l
'mo' The rise of the noise removal difference zero crossing number is exceeded for the first time 9 times.

2 “hour”°]’・“57) Next, use the D-5 plane, 〇-△ year plane, and N-D plane to calculate 17
Species phonemes /a/, /ka/, 15a/, /La/.

/na/, /ha/, /ma/, /ya/, /ra/
, /wa/, /pa/.

/a/, /za/, /ga/, /cla/, /ba/
, /N/.

On the 0-△ plane /a/, /za/, /ga/, /d
a/.

/b a/ is not treated, and on the N-D plane, /k a /
,/ta/.

/pa/, /ya/, /ra/, /wa/, and /N/ are not handled.

The reason for this is that it has been experimentally confirmed that these phonemes have the following distribution. The distribution of /a/ on the 〇-△ plane is the same as /Sa/, /za/ , /
The distribution of ga/, /da/, and /ba/ is similar to /na/, but the distribution is wide. /ka/ on the N-D plane,
The distribution of /la/, /pa/ is between /a/ and /ha/, /ya/, /ra/, /wa
The distribution of / is the same as /a/.

Yotte, 0-△ plane Te/a/, /za/, /ga/,
/da//l) a4 handling and /k in the N-D plane
a/I/La/;'/pa/, /ya/, /ra/, /
Handling wa/ worsens the identification results. Also, /N/ cannot be plotted on the ND plane because there is no oral cavity output.

For the above reasons, only limited phonemes are handled on each plane.

In order to compensate for this restriction, the algorithm described below performs identification on each plane multiple times.

FIG. 11 is an identification flowchart. In step 5, /N/ is separated and identified on the 〇-△ plane. This is because the distribution of /N/ on the O-Δ plane is very prominent, and it is appropriate to first separate it from other phonemes. 5
In step 2, on the N-D plane -/sa/, /a/
, /za/, (/na/, /ma/, /ga/, /da
/, /ba/) are separated and identified. In 5tep3, on the 0-△ plane again (/ya/, /ra/, /wa/
) is separated and identified. 5tep4 Teha D-5
On the plane, (/a/, /h a/ ) and ('/ka/
, /la/, /pa/) and 5 steps.
5, discrimination between /a/ and /ha/ is performed again on the ND plane. The reason why the five-stage configuration is adopted in the above-mentioned five-stage processing is that a method is adopted in which phonemes that are significantly separated from other phoneme groups on each plane are first separated and identified. On the other hand, cases where the following vowel is /e/ or 10/ can also be identified in principle using the flowchart of FIG. 11. However, when the following vowel is /1/, /U/, the affricate /l i/,
/lsu/ is separated in 5tep2. Note that /n]i/ on the 0-△ plane.

The characteristics of /mu/, /yu/, etc. are unclear, and if the following vowel is /V or /u/, a post-minute test 3'NJ is required.

In the recognition of the above-mentioned speech, more accurate recognition can be achieved by providing a detector using a camera or the like that detects mouth movements and performing detection using this detector and the detection device of the present invention.

As described above, in this invention, two audio reproducers are provided at a predetermined interval at the tip of an arm fixed to the head, and one audio reproducer detects nasal cavity output and the other detects mouth+r<g output. Because of this, it can detect audio more accurately than a single microphone input device, making it very effective as part of a computer's audio input device.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram showing an example of the present invention, Fig. 2 is a block diagram of the system using this invention;
The force is a graph showing time and energy distribution, Figure 4 is D
-8- A graph showing a flat curve, Figure 5 is a graph showing the energy curve passing through the pharyngeal cavity of /ma /, %6 is a graph showing an O-△ flat curve, Figures 7 and 8 are Figure 4 Figure 6 is a corrected graph, Figure 9 is a graph showing an example of correction, and Figures 10 (a) to (e) are the following vowels /2/ for 1° male.
, /V, fi/, /e/, 10/ in each monosyllable,
A graph showing an N-D plane in which the noise removal difference zero crossing number is plotted on the vertical axis and the rise time difference between the noise removal difference zero crossing number shown in Equation 12 and NEn is plotted on the horizontal axis. FIG. 11 shows an example of speech recognition. It is a flowchart. 2... Head arm, 3m, 3n... Microphone, 4.
...Support arm Figure 3 (a) Figure 3 (b) Figure 3 (i)
Figure 3 ('') Figure 3 Busy) 3
Figure (d) Figure 3 (9) Figure 3 (h) Figure 3 (k)
Figure 3 (J) Time (msec)
, R-fmsec) (En/Eo). (En/Eo). Procedural amendment (View 1. Indication of the case 1982 Patent Application No. 72420 2, Name of the invention Audio wave detection device 3, Person making the amendment Relationship to the case Patent applicant address 4-1 Iguchicho, Takamatsu City Name ( Name) Hiroshi Jinnai Contents of the amendment as shown in the attached sheet 1. "Regularized" on page 3, line 17 of the specification is amended to "normalized". 2. Same as above, page 6, line 9. Observation value En (in) CL of the eye
) to [observation value En(in)(1) + Em(1
n)Li) to ”. 3. Correct "same cylindrical shape" in the first line of page 7 of the above to "cylindrical shape" and omit it. 4. [En(t)/Eo(t
'River 'En(t)/Eo(t) j K Correction L
4. 5. Lines 5 to 15 of page 9 of the above are amended as follows. The zero-crossover number for noise removal is defined by the following equation (8). Here, for each monosyllable, we pay attention to the time difference between the rise time of the noise-removed difference zero-crossover number and the rise time of NE, and calculate the difference as shown in Figure 10. The noise removal difference zero crossover number (N, R, -D, Z
. 9) is defined as 0 Next, the D-5 plane, 0-△ plane, and N-D plane are defined as "7,
Same as above, page 10, line 9, same page, line 11, page 1, page 16
line, page 11, line 1 and page 14, line ``/a/
" is corrected to "/fa/" and deleted. 8. Change r/l i/j in the 7th line of the 12th member to “/lJ
Correct it to 'I/' and cut it out. 9, "Formula 12" on page 13, line 14 of "Ibid." 2 is changed to [Formula (9
)”. Procedural amendment (voluntary) February 10, 1980 1. Indication of case 1988 Patent Application No. 72420 2. Title of invention (old) Sound wave detection device (new) Sound wave detection method 3. Make amendments. Patent applicant address: 4-1, Iguchi-cho, Takamatsu-shi Name: Norihiro Kannai 5゜Showa year, month, day (shipment date) Contents of the amendment (1) as attached (1) Full text of Specification 1 as attached I'll correct it. (2) Drawings 5i, 7, and 8 attached to the application. Figures 9 and 10 have all been deleted, Figure 11 has been corrected to Figure 10 as if the attached drawings were not copied, and Figures 5 and 7 have been added to the Kasuri lattice as shown in the attached sheet. Added figures 8, 9 and 11. (3) The name of the invention in the application is amended to "sound wave detection method." Description 1, Title of the invention Sound wave detection method 2, Claims Detecting the sound output from the oral cavity and simultaneously detecting the sound output from the oral cavity, and determining the emitted sound wave based on the output values of both types. An audio wave detection method characterized by: 3. Detailed Description of the Invention The present invention relates to a voice detection method for detecting the voice of a person (6). With the development of computers, voice input devices have been proposed. This voice input device stores the characteristics of voice waves in advance in a voice storage unit, and compares the characteristics of newly input voice waves with the values stored in the voice storage unit to recognize the input voice. . However, since conventional speech recognition identifies and recognizes phonemes using speech waves that are a combination of nasal and oral cavity outputs, it is not fully satisfactory in terms of recognition rate and recognition time. divided into those facing the nasal cavity and those facing the oral cavity,
Nasal cavity 9 It is output from the nasal cavity and lips after being modified by transmission characteristics based on differences in the shape of the oral cavity. At this time, the shape of the nasal cavity remains unchanged, the velum of the mouth moves, the shape of the oral cavity becomes hollow due to the movement of teeth and tongue, and the lips open and close. Therefore, the vocal cord vibration waves generated from the vocal cords and divided into the oral cavity and nasal cavity are modified by the woodcutter due to the difference in the shape of the passage from the nostrils to the lips, so the output waveforms from the nostrils and lips of the two are clearly different. Become something. Furthermore, even when producing an unvoiced sound, the nasal cavity output is extremely small, only the oral cavity output exists, and the output waveforms of both are clearly different. This invention was made by focusing on the fact that the speech waves emitted from the oral cavity and the nasal cavity exhibit clearly different waveforms, and aims to make recognition of phonemes extremely easy. In order to achieve this objective, the sound wave detection method of the present invention detects the sound output from the oral cavity at the same time as the sound output from the nasal passages, and detects the sound waves emitted based on the values of both sides. It was decided to do so. According to this method, a threshold value of a parameter serving as a recognition standard is determined through a preliminary experiment, and the voice uttered by an unspecified person is recognized by comparing it with the determined threshold value. Therefore, according to the present invention, since the speech is recognized based on two clearly different waveforms of the nasal and oral cavity outputs, phoneme recognition is extremely easy, and when computer processing is performed, the recognition rate and At the time of recognition, the memory capacity is significantly improved. Embodiments of the present invention will be described below with reference to the accompanying drawings. As shown in FIG. 1, from a head arm 2 fixed to a person's head 1, an arm 4 supporting microphones 3 m and 3 n at its tip extends in front of the face in an adjustable manner. The microphone 3n faces the nostril to detect the nasal cavity output of the sound, and the microphone 3m faces the mouth to detect the oral cavity output of the sound. Note that it is desirable that either one of the microphones 3n and 3m is provided so that its position ss can be adjusted. Further, it is desirable to provide a shielding plate 5 between the microphones 3m and 3n to separate the nasal cavity output and the oral cavity output, so that both outputs are inputted to each microphone 3n and 3m without being mixed. The signals from each of the microphones 3 m and 3 n are passed through a switch 6 such as a multiplexer and sent to A as shown in the second section 1.
The signal is input to the D f converter 7, converted into a digital signal, and input to the computer 8. The computer 8 performs recognition processing of the emitted voice based on the outputs of both microphones 3 m and 3 n. This recognition process can be performed by various means, for example as follows. This means is based on the energy curve i13)lNm of the audio output detected by the microphone 3 +n L3n,
Based on Nn, the following 0) 1), S, (En/Eo
). ,,Δ(E2n/Eo). Calculate the parameters CR, DT, and use each parameter to obtain D-5, (En/F, o).・-Δ(E
n 7' J+, 0) plane (0-Δ plane), CR-D
The T plane (IN-D plane [fn) is obtained, and vocalizations are identified based on each plane. D, rise time difference S between n and Nm: ,p+slope ratio En at the rise of En and Nm
: Energy passing through the nasal cavity EO: Energy passing through the pharyngeal cavity (En/Eo) 0 : En /, E □ Curve opening starting point value Δ (En/Eo): Maximum slope value C of the En/Eo curve
k: noise removal differential zero crossing number DT: noise removal differential zero crossing number and the rise time difference between NEn, and the creation of each of the planes and the phoneme identification based thereon will be described. 1) o-5+ side Figure 3fa) -fI is the voice tn/A/, ka)/k
a/, m, /Sa/l)/la/, +su+/na/
, ri/ha/, f71/ma/, tri/ya/
, (U) / r' a 7, i'71 / w a
/, H/pa/, fn)/N/ my
3m. Energy curve N・Em obtained by detection with 3n, NE
n, and is an energy time change curve obtained by normalizing each observed sound wave energy by its maximum value. In this energy curve, /a/, /h2L/te, N, n, and m rise simultaneously, and N, n during utterance.
and Nm have the same change. /ka/, /La/,
/l) a/te has a peak due to the bursting air flow when Nm rises, and Nm rises 9 times more than Nn. /sa
A small value (arrow) appears in Rm in the /S/ section. /na/, /ma/TEHA NEn rises more firmly than NEm, and at the same time as N- starts increasing, NEn starts deriving. For /ya/, /ra/, and /wa/, Nn and Nm rise almost in the same direction, or the slope at the time of rise is larger for Nn than for Nm. The oral output for //N/ is extremely small, and the energy curve of indoor noise appears for N''In. The delay defined by the following equation f1+, (2+) is the parameter representing the characteristics of the energy curve of each phoneme. Calculate the time period and slope ratio S. D=1 −1 no mo ......(1)
However, 'no' mo is the time when N n and N m each exceed the 5% point of the maximum value for the first time, [m3 (ho). 1) is the rise time difference between Nn and NEm, and equation (2) is N
It represents the ratio of the slopes of Nn and Nm at the rise of m. Figure 4 excludes /sa/ and ha/ in Figure $3 <10kO
The two parameters of the M clause are calculated and plotted on the D-8+ surface with the initial letter of the vocalization, for example /la/, as T. The speech samples are isolated utterances by 10 men. In addition, in /SiL/, the variation in NEm value is large in the /S/ interval, making detection of L city 0 unstable, and /N/
This case was excluded because the oral cavity output was extremely small and tmo could not be determined. In the drawing, if S ) 3.0, then s = 3.0.
It is plotted at the 0 position. According to this figure, when /a/ and /ha/ are delayed, 1'iJ
D is small, and the slope ratio S is distributed around 1 and 0. /
ka/. In /la/, /Pa/, D ”> Q and S are small. Also, /na/. /ma is D<O, S<0. /Y ” + /
For ra/ and /wa/, D is small and S is larger than other phoneme groups. There are speech samples in which this phoneme is slightly delayed (D<O) from the onset of NEm or NEn, but they are
The outlines of NEn and NEm had almost the same shape as in Figure 3, or the first peak of the N''n curve was hard, and the slope of Nn was calculated near the top of that peak, so the slope ratio S was slightly different. It can be said that it has become smaller (arrow in Figure 4). From the above, it can be understood that by finding D, 5, each phoneme can be classified into several types of groups. 11) 0-Δ plane As shown in Figure 1 Then, the air key passing through the pharyngeal cavity 9 is E. (0, nasal cavity 10 to 0) P!: 1' 10) Out crab, r-
Enfjl, % I) Also, Mike 3n
, 3m observes energy as E, (in]cl, E
11 mo with m(in]l, observation value En(in)
Estimate the time sharpening curve of Il from Em(in]cl. Eo[Ll = Eo[Ll = Enftl + Em(Ll
--f3+Also, if Cn・Cm is the ratio of radiated energy entering each microphone, then from equation (3i and equation (4)), CnEo(Ll-En(in) (t) + (Cn/C
ITl)Errl(in)it)--f51 is obtained. Here, the problem is the calculation of Co/Cm.For example, one microphone is placed in the opening at one end of a cylinder, and the opening at the other end covers the mouth and nose. At the same time, Em and En are detected by the heavy membrane shown in Fig. 1, and Eo=EITl+E
It is sufficient to calculate Cn''m which is n. Since Cm and On change depending on the positions of the microphones 3m and 3n, it is preferable to fix them and compare E.+Em'En with the average value of a plurality of times. Based on Cn/Cm obtained in this way, the following equation (6) is obtained. Figure 5 shows En/E of the above equation (6) for the vocal sound /a/... 〇This E shows the time change curve
The following equation is defined as a parameter representing the characteristics of the curve of n/F, o. (En/Eo)o=En(Lo)/Eo(to) +
'++++ i7] and Max means the maximum value inside the parentheses. Also, 0 is Eo or 15% of the maximum value
The time when E' crosses the point for the first time, E' is an arbitrary time from t In addition, Equation (8) represents the maximum slope of till k. Figure 6 shows the above two parameters of the 12 types of monosyllables uttered in isolation by the 10 men shown in Figure @5. It is plotted by the initial letter of the vocalization on the -Δ plane. (6th
In the figure, only five Jingei samples are plotted to make the diagram easier to read, but the other five samples also have the same core distribution. ) Due to this reason, the double sound of 111Mf “0 is (/a/, /k
a/. /sa/, /La/, /]1a/, /Pa/ ), (
It is classified into four groups: /na/l/ma/), (/ya/. 7r al, /wa/), and (/N/). 1.1'o, and (7) flat m upper Teha 4a/, /za/
, /ga/, /da/. /h al is not handled, or the distribution of φ/ is the same as /Sλ/, /za/, /ga/, /da/, /ba/
0) 'A cloth is similar to 10a/, but the distribution is wide and difficult to classify. 111) N-D plane This plane identifies t-m fricatives such as yS/, /z/, and affricates such as 7tJiot/lsu/.
One of its parameters is the number of zero-crossings for noise removal difference (N
o1se rejected diff, erenti
al zero crossing rate). For example, /sulo) oral cavity output audio waveform (
Here, the oral cavity output audio signal at a certain point is (Xi)
Then, the noise removal differential zero crossing number CR is defined by the following equation (9). $8 Figure tal to mountain /u/ , /s u/ , /z
u/ , /l s u/ , /hu/ +/nu/Q
It shows the relationship between the CR of the pronunciation and time. Next, as another /se y meter, use the following formula +1
The rise time difference D T (Del
ay time). Here, each rising edge of phEn, NF, and m is t.
no, tmo. 11. The rise of the sound removal difference zero crossing number is assumed to be time 7 () when the number of crossings exceeds chi9 times for the first time. Figure 8 (a
1bj change curves Nn and Nm during normalization of the output power of the birch cavity 1 are shown in Figure 9.
−C・R,] (CR) is the noise removal difference zero crossover number C on the vertical axis.
The rise time difference between R and Nn (1) clay cime) is plotted on the 117j axis number Ko 1 Otsu song (N-D plane) for each early spring festival by the following mother's meaning by the initial letter of the vocalization sound. . According to this diagram, the monosyllables /S/, 4/, /z/, /h/
, (/nX7m/, , /g/, /d/, /b/) and O
・I can confirm that it can be classified into vowels. In this figure, /ka/, /[al, /Pa/
, /γa/. Doesn't it handle /ra/, /wa/, /N/?1/k
al,/tal. /pa/(7) The distribution is /g/, /h L/(') ('
Located in a, +, /ya/, /r2L/. /Wa/? 7) The distribution is the same as /a/, which is inconvenient for classification, and /N/ has no oral output, so N −
This is because it cannot be plotted on the D plane. The method for creating each plane has been described above, and an example of an algorithm for phoneme recognition using these planes will be shown below. An example of this algorithm is the phonemes /a/, /ka/, /
sa4/[a/, /na/, /ha/, 7m2L/, /
ya/l/ra/, /Wa/l/Pa/, /, /a//
I/za/, /ga/', /da/, /ha/, /N/
This is done according to the identification flowchart shown in FIG. 5tep 1 Separate and identify //N/ on the 0-Δ plane. This is because the distribution of //N/ on the 0-Δ plane is very remarkable, and it is appropriate to first separate it from other phonemes. te7 s a/+
/fi/ + /z a/ +(/ma/, /na/,
/ga/, /dA/, /ba/) (7) Separate and identify the 4 groups. 5 step 3, again on the 〇-∆ plane Te(/'ja/
, /ra/. /wa/) is separated and identified. 5tep 4, on the D-5 plane (/a/, /h
a/ ) and (/ka/, /la/, /pa/) are separated. In step 5, /a/ and /ha/c) are again identified on the N-I) plane. The 17 types of phonemes are classified into 9 groups through the above 5-step processing. The reason for this five-stage configuration is that a method is used to first separate and identify phonemes that are significantly separated from other phoneme groups on each plane. On the other hand, the following vowels /e/ and 10/ are also identified using the flowchart shown in FIG. 10 in principle. In addition, the following vowels are /i/, /u/, and the affricate /l'fi%/lsu/ is 5LeP.
Separate by 2. Within each phonological group classified into multiple groups in this way,
Each phoneme is identified using a conventional and well-known recognition method, such as a recognition method based on the centroid frequency, peak frequency, and cylindrical wave number of the spectrum and their temporal changes.
make the final judgment. As shown in , the energy curve i, 9B NEm , N
For En, connect the 35% point and 15% point of the maximum value with a straight line, and the intersection of this straight line and the time axis with Lno + (Lm
After making corrections such as o), the D-5 plane 〇-Δ plane,
What is necessary is to create an ND plane. Note that in the voice recognition described above, mouth movements can be accurately identified. 4. Brief description of the drawings Figure 1 is an explanatory diagram showing an example of this invention, Figure 2 is a control block diagram using this invention, and Figure 3 is a graph showing time and energy distribution. , Figure 4 is 1)-5
Graph showing the plane, Figure 5 tal-+1 is E, /Eo
Figure 6 is a graph showing the O-Δ plane, Figure 7 is the oral output speech waveform of /Su/, and Figure 8 ia) to if) are the normal nasal cavity 1 oral cavity output energy. Figure 9 (a) - tel is N-
The graph showing the D plane, FIG. 10 is a flowchart showing an example of voice recognition, and the eleventh figure is a graph showing an example of correction. 3m, 3n...Microphone, 4...Support arm. Patent applicant Norihiro Jinnai Agent Bunji Kamata Figure 5 (a) Figure 5 (9) Figure 8 (a)
Figure 8(b) Figure 8-(c) Figure 8(d)
Figure 8 (e) Figure 8 (f) Figure 10 Procedural amendment (master) 1. Indication of the case 1982 Patent Application No. 72420 2. Name of the invention Sound wave detection method 3. Person making the amendment Case Relationship with Patent Applicant Address: 4-1 Iguchi-cho, Takamatsu City Name: Norihiro Kannai Address: 〒542 Spikyanai〒〒 π Usuhika de Hakumoe Akuta ■yo: Hiyo Hikaya β generation: 5゜1974 Month, Day (Delivery date) Contents of amendment ■ "Nasal cavity" in line 4 of page 2 of the full text of the amended specification based on the procedural amendment dated February 10, 1980 will be corrected to "nostril" (IC). 2 r /xi ten l/ j and "/X17" on page 11, line 12 of the same ministry, respectively, are replaced with "IX1 October" and rlXi?
Jf/m correction. ;3 Change [/xi-1/J to rlxi-1 on line 13 of the same page
White 11 is correct in 1J.

Claims (1)

[Claims] A sound wave detection device characterized in that two sound regenerators are provided at a predetermined interval at the tip of an arm fixed to the head, and one sound regenerator detects a nasal output and the other one detects an oral cavity output. Device.