JPS619696A - Voice recognition equipment - 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] Industrial applications The present invention relates to a voice recognition device for recognizing voice.

BACKGROUND TECHNOLOGY AND THEIR PROBLEMS Conventionally, there has been a sound system that deals with variations in speech rate.
For example, an apparatus has been proposed that performs P matching processing, as disclosed in Japanese Patent Laid-Open No. 50-96104.

First, a speech recognition device that performs speech recognition using this DP matching process will be described.

In FIG. 1, (1) is a microphone as an audio signal input section, and the audio signal from this microphone (1) is supplied to an acoustic analysis section (2), and the acoustic analysis section ff1
i (21, the acoustic parameter time series Pi(nl) is obtained.

In this acoustic analysis section (2), for example, the rectified and smoothed output of the bandpass filter bank is calculated as the acoustic parameter time series Pi.
(nl (i = 1....+I; I is the number of channels of the bandpass filter bank, n-1,...,
NUN is the number of frames extracted by sound PR section determination. ) is obtained as

The acoustic parameter time series Pifn of this acoustic analysis section (2)
) is stored in the standard Bakun memo 1/4) for each recognition target word in the registration mode by the mode F changeover switch (3), and is supplied to one end of the DP matching distance calculation unit (5) in the recognition mode. Further, in this recognition mode, the standard pattern stored in the standard pattern memory (4) is supplied to the other end of the DP matching distance calculation section (5).

This DP matching distance needle section 5) performs a DPP matching distance calculation process between the input pattern consisting of the acoustic parameter time series Pi (nl) of the human input voice at that time and the standard pattern of the standard pattern memo January 4. , this D
A distance signal indicating the DPP matching distance from the PP matching distance calculating section (5) is supplied to the minimum distance determining section (6), and the minimum distance determining section (6) calculates the DPP matching distance for the input pattern.
The standard pattern with the minimum matching distance is determined, and from this determination result, the recognition result indicating the input voice is output to the output terminal (7
) can be obtained.

Incidentally, the number N of frames of the standard pattern stored in the standard pattern memo (January 4) generally varies depending on variations in speaking speed and differences in word length. The D P matching process performs time axis normalization to deal with variations in speaking speed and differences in word length.

This DPP matching process will be explained below. Here, for simplicity, the dimension corresponding to the frequency axis direction i of the acoustic parameter time series P i (nl is omitted, and the parameter time series of the standard pattern is expressed as bl, . . . ).

bN・The parameter time series of the input pattern is a□.

. . +8M, the DPP matching process in the case of a DP-bus with fixed end points will be explained.

Figure 2 shows a conceptual diagram of the DPP matching process, with input parameters (M = 19) arranged on the horizontal axis and standard parameters (N = 12) arranged on the vertical axis. , N) There are MXN points in the grid plane,
One distance corresponds to each point. For example a3 and b5
The distance corresponds to the intersection of the straight line extending vertically from a3 and the straight line extending horizontally from b5. in this case,
For example, if we take the Chebyshev distance as the distance, a3 and b
The Chebyshev distance d (3, 5) with respect to 5 is (in this case, I=1 because the dimension corresponding to the frequency axis direction i is omitted). Then, as a DP-bus with fixed end points, this grid point (m, n) is
The left grid point (ml, n
), if only three lattice points (m-1, n-1) on the diagonal left side (m-1, n-1) and grid points (m, n-1) on the g-side are allowed, then the starting point, that is, the connection between al and bl. Chebyshev distance 1tll
t l) Starting from point 0, which shows 11, select 3 directions 2I as the bus (route), and take a bus to point ◎, which shows the Chebyshev distance d (M, N) from the end point, that is, aM,
Find the one that minimizes the sum of the distances of each passing grid point, and divide this sum of distances by (M+N-1), which is the sum of the number of input parameters M and the number of standard parameters N, minus the value 1. The obtained result is subjected to DPP matching between the parameter time series al+...+8M of the input pattern and the parameter time series bx,...rbN of the standard pattern. The initial condition and recurrence formula showing such processing are expressed as the initial condition g (1,,1) -d (1,1) recurrence formula, and from this, the DP matching distance 11tD (A,
B) is D (A,,B)=g (M,N)/(M+N-
1) (The reason why g (M, N) is divided by (M+N-1) is to correct the difference in distance value due to the difference in the number of frames N of the standard pattern.)

Through such processing, when there are five standard patterns, five DPP matching distances for the input pattern are obtained, and the standard pattern with the minimum distance among these five DP matching distances is determined as the recognition result.

As described above, the speech recognition device using P matching processing can deal with variations in speaking speed and differences in word length, that is, perform speech recognition with time axis normalization.

However, when speech recognition is performed using P matching processing like this, the stationary parts of the speech are largely reflected in the DPP matching distance, and it is easy to misrecognize between words that are partially similar. It became clear that

That is, the acoustic parameter time series Pifnj can be considered to draw a trajectory in the parameter space. Actually, since the parameters of each frame n correspond to one point in the parameter space, if the points are connected by curves in the time series direction, one locus from the starting point to the ending point can be considered.

For example, when two types of words ""SAN" and "HAI" are registered, the respective standard patterns A' and B' are "S", "A", "N", and "H" as shown in Figure 3. , "A",
Draw a trajectory that passes through each phonetic area of ``■''. Then, when ``SAN'' is uttered in the recognition mode, overall, there are very few similarities between standard pattern B' and input pattern A. However, the “SAN” of this input pattern A is
The “A” part of “” is the “SAN” of standard pattern A’.
"A" of "HAI" of standard pattern B' from part "A"
, and there are cases where there are many points in that part (quasi-stationary part).

Here, as shown in FIG. 3, the parameters of the input pattern A are generally similar to the parameters of the standard pattern A',
A case will be described using a one-dimensional parameter as an example, where DP matching processing causes erroneous recognition when the pattern is partially similar to the variation of standard pattern B'. In this case, the situation shown in FIG. 3, that is, the input pattern A as shown in FIG. 4 as a one-dimensional parameter time series similar to the relationship between partially similar words; 2.4.6.8 .8.8.8.6.4
．． 4.4.6°8 and 1. Standard pattern A' as shown in FIG. 5; 3,5°7, 9.9.9.9.7.5.5.7.
9, and a standard pattern B' as shown in FIG. 6, 7.
6.6.8.8.8.8.6°4, 4.4.

As is clear from the patterns shown in FIGS. 4 to 6, input pattern A is a pattern that is desired to be determined as standard pattern A'.

However, when calculating the DP matching distance of standard patterns A' and B' with respect to input pattern A, it is shown that input pattern A is close to standard pattern B'.

That is, the DP of standard pattern A' for input pattern A
Similar to FIG. 2, as for matching processing, as shown in FIG. 7, the horizontal axis represents the parameter time series i of input pattern A. 2.4.
6.8.8.8.8.6.4.4.4.6゜8 arranged,
The vertical axis is the parameter time series of standard pattern A', 3.5
．． 7.9.9.9.9.7.5.5.7.9 are arranged, and each parameter of standard pattern A' is calculated for each parameter of input pattern A corresponding to the intersection in the grid plane. Find the Chebyshev distance. Then, the Chebyshev distance 1i1td'(1,1
) -1(7) point as the starting point, the 13th parameter 8 of the parameter time series to the input pattern, and the 12th parameter 9 of the parameter time series of the standard pattern A'
Chebyshev distance 111d (13,'12) -1
Set the point as the end point, and as in the case of Figure 2 as the DP-path, take the point to the left of that arbitrary point, the point below it, and the point on the diagonal left + side as the previous state for the point of negligence. (This path is indicated by a solid arrow), the points on the path are d (1, 1) - d (2, 2) - d (3, 3) - d
(4,4) −d (,5,5) −d(6,6) −
d (7, 7) - d (8, 8) - d (9, 9
>-d (10,10) -d (IL 10
)-d(12,10)-cl(13,11)
-d (13, 12), the total distance is 24, and this DP matching distance D (A,
A') is 1.

On the other hand, DP of standard pattern B' for input pattern A
The matching process is performed as shown in FIG. 8, as in the case of FIG. 7 described above. That is, input pattern A's (IM parameters; 2.4.6.8.8.8°8, 6.4.
Find the Chebyshev distance of standard pattern B' for 4.4.6.8 (fixed parameter i 7.6.6.8.8.8. '8.6°4, 4.4, and calculate DP- As a path °ζIf we are allowed to take the points to the left, the points below, and the points diagonally to the lower left of any point as the previous state for that point (this path is shown by a solid arrow), The points on the path are d (1,1)-d (2,2)-d (3,3,)
-d (4,4) -d (5,5) -d (6,
6)-d (7,7)-d (8,8)-d(9,
9) -d (10,10) -d (11,11)
-d (12, 11) -d (13, 11), and the sum of their distances is 15, and this DP matching distance m1tD (A, B') is 0.65. , As is clear from the result of using this DP-path as a three-directional force, the input pattern A is determined to be the standard pattern B' whose DP matching distance is small, and the result to be determined cannot be obtained. In this way, in the DP matching process, it is easy to misrecognize words that are partially similar.

In addition, in the DP matching process, as mentioned above, the number of frames N of the standard pattern is undefined, and it is necessary to perform the DP matching process on all standard patterns for the input pattern, and as the number of words increases, the amount of calculation increases accordingly. This has caused problems in terms of the storage capacity and amount of calculation required for standard pattern memory.

For this reason, there are relatively few misrecognitions even between words that are partially similar, and the storage capacity and amount of calculation for processing of the standard pattern memo (January 4) is relatively small. A voice recognition device and a snake as shown in FIG. 9 have been considered.

In FIG. 9, (11 indicates a microphone as an audio signal input section, the audio signal from this microphone Tl) is supplied to the amplifier (8) of the acoustic analysis section (2), and the audio signal of this amplifier (8) is The cutoff frequency 5.5KHz
The digital audio signal from the A/D converter αψ is supplied to a 12-bit A/D converter aω with a sampling frequency of 12.5 KHz through a Cubas filter (9) of 15 channels.
, (lls) , ”..., (llo).

This 15-channel dental bandpass filter bank (llA), (11B), ”, (l
lo) is composed of, for example, a Butterworth fourth-order digital filter, and the bands from 250 Hz to 5.5 KHz are distributed at equal intervals on the logarithmic axis. And each digital bus fill (118), (ll
s).

..., (llo) output signal through a 15-channel rectifier (12A) + (12B) + ....+
(12o) respectively, and these rectifiers (12^
), (12s),...

The square output of (12o) is passed through a 15-channel digital low-pass filter (13A), (13g),
..., (13o), respectively. These digital low-pass filters (13A), (13s),
”, (13o) is constituted by an FIR (finite impulse response type) low-pass filter with a cutoff frequency of 52.88Z.

and each digital low-pass filter (llA).

The output signals of (13B), . . . , (13o) are supplied to a sampler (14) with a sampling period of 5.12 m5.

This sampler (14) allows digital low-pass filters (13A > , (13s) , ..., (
The output signal of 13o) is sampled every frame period of 5.12m5, and the sampling signal of this sampler (14) is supplied to the music information normalizer (15). This sound source information l111. The 'iE normalizer (15) removes differences in vocal cord sound source characteristics depending on the speaker of the speech to be recognized.

That is, the sampling signal A11n) (i = 1...) supplied from the sampler (14) every frame period.
....

15Hn: frame number) for Ai(nl=log(Ai(nl+B)
'...if) is performed. This (1
), B is set to a value such that the noise level is hidden by the bias. Then, the vocal cord sound source characteristics are yi=a−
It is approximated by the formula 1+b. The counts of a and b are determined by the following equation.

(N=15) ... (2) N (N-1) (N=15) ... (3) Then ζ, if the normalized parameter of the sound source is Pi (nl), when aH < 0 Parameter Pi (nl is Pi
Tn) −AiTnl −(a(ni i +bTnl)
... is expressed as +41.

Also, a (only normalizes the level when nl is 0,
The parameter Pi(n) is expressed as... (5).

Through such processing, the normalized parameters Pifn) of the vocal cord sound source characteristics are stored in the vocal interval parameter memory (16).
supply to. Parameter memory within this voice section (16)
is a normalized parameter P of the vocal cord sound source characteristics in response to a voice interval determination signal from a voice interval determination unit (17) to be described later.
ifn) is stored for each audio section.

On the other hand, the digital audio signal from the A/D converter (10) is supplied to a zero cross counter (18) and a power calculator (19) of the audio section determining section (17), respectively. This zero cross counter (18) counts the number of zero crosses of the digital audio signal at 64 points in that section every 5.12 m5, and supplies the count value to the first input terminal of the audio section determiner (20). Also, the power calculator (19) calculates the power of the digital audio signal in that section every 5.12 m5, that is, the sum of squares, and calculates the power signal without calculating the power within the section and sends the power signal to the second voice section determiner (20). Supplied to the input terminal of Furthermore, the sound source normalization information a (nl and b) of the sound source information normalizer (15)
tn+ is supplied to the third input of the speech segment determiner (20). Then, in the voice section determiner (20), the number of zero crossings, the power within the section, and the sound source normalization information a fn
l, b (n) are processed in a composite manner, and a process for determining silence, unvoiced sound, and voiced sound is performed to determine a voice section. A voice interval determination signal indicating the voice interval of the voice interval determiner (20) is supplied to the voice interval parameter memory (16) as an output of the voice interval determination unit (17).

The normalized acoustic parameters Pifnl of the vocal cord sound source characteristics are stored in this intra-speech-segment parameter memory (16) for each speech period and are SAT (Normal) in the chronological direction.
lization Along Trajectory
) is supplied to the processing section (21). This NAT processing unit (21
) is the acoustic parameter time series pitn as NAT processing.
A trajectory in the parameter space is estimated from l by linear approximation, and a new acoustic parameter time series O1 (@) is formed by linear interpolation along this trajectory.

Here, this NAT processing section (21) will be further explained. Acoustic parameter time series Pi(n) (i = 1.
...+I i n-1+ ..., N) draws a point sequence in the parameter space. FIG. 10 shows an example of a point sequence distributed in a two-dimensional parameter space. As shown in FIG. 10, the point sequence of the non-stationary part of the voice is roughly distributed, and the quasi-constant part is densely distributed. This is clear from the fact that if it is completely stationary, the parameters will not change, and in that case the point sequence will remain in the parameter space.

FIG. 11 shows an example in which a locus consisting of a smooth curve is estimated and drawn on a series of points as shown in FIG. If a trajectory can be estimated for a sequence of points as shown in FIG. 11, it can be considered that the trajectory remains almost unchanged with respect to variations in speech rate. This is because most of the differences in time length due to variations in speech rate are due to the temporal expansion and contraction of the quasi-stationary part (in the dot sequence shown in Figure 10, this corresponds to the difference in the density of the dot sequence of the quasi-stationary part). This is because it is thought that the influence of the time length of the unsteady part is small.

The NAT processing unit (21) performs time axis normalization by focusing on the invariance of the trajectory with respect to such variations in speech rate.

That is, first, a trajectory is estimated as a continuous curve from the start point Pi (1) to the end point Pi (9) for the acoustic parameter time series Pi (nl), and the curve representing this trajectory is defined as 'P'i[s )
(0≦S≦S). In this case, Pi(o
l = Pifll.

It suffices if it passes approximately through the entire point sequence.

Second, obtain the length SL of the trajectory from the estimated P1(sl), and resample a new point sequence at a constant length along the trajectory as shown by the circle in Fig. 12. For example, sample at point M. In this case, a fixed length, i.e., the number sampling interval T
The trajectory is resampled based on =SL/(M-1). This resampled point sequence is converted into the old fml (
i=L-...,I; m=1. -...-, M) The new parameter time series Qi (ml) obtained in this way has the basic information of the trajectory and is a parameter that is almost invariant to variations in the speech rate. , the new parameter time series old (ml is the parameter time series that has been time-axis normalized).

For this kind of processing, parameter memo 'J
The acoustic parameter time series Pifnl of (16) is supplied to the trajectory length calculator (22). This trajectory length calculator (22)
calculates the length of the trajectory drawn by the acoustic parameter time series WtlPi(n) in its parameter space based on a straight line, that is, the trajectory length. In this case, as the distance between the I-dimensional vectors a1 and b□, for example, the Euclidean distance 1i
If lltD (ai, N) is taken, ■... (6). Note that this distance may be Chebyshev distance, square distance, or the like. Therefore, ■-dimensional acoustic parameter time series Pil11> (1= L..., l
;n-1,...,N), the distance between adjacent parameters in the time series direction when 11 trajectories are determined by linear approximation is 1i11iS (nl is 5(nl - D (Pi (nlx),
Pi(nl) (n = 1......,
N-1)・・・・(7) Then, the first parameter Pi (11th to nth parameter PiC
The distance st, tn> to nl is expressed as. In addition, 5L (11 = 0.Furthermore, the trajectory length SL is expressed as.The trajectory length calculator (22) uses this equation (7),
The signal processing shown in equations (8) and (9) is performed.

A trajectory length signal indicating the trajectory length SL of this trajectory length calculator (22) is supplied to an interpolation interval calculator (23). This interpolation interval calculator (23) calculates a constant length sampling interval T for resampling a new point sequence by linear interpolation along the locus. In this case, if resampling is performed at M points, the resampling interval T is T=SL/ (M-1) ・-・QOI
It is expressed as The interpolation interval calculator (23) calculates this (1(1
The signal processing shown in equation 1 is performed.

This interpolation interval calculator (23) resamples the interval signal indicating the sampling interval T to the interpolation point extractor (24)
Parameter memory within the speech interval as well as supplying to one end
The acoustic parameter time series 1'1(n) of (16) is supplied to the other end of the interpolation point extractor (24). This interpolation point extractor (
24) resamples a new point sequence at the resampling interval T along the trajectory of the acoustic parameter time series Pi[nl in its parameter space, for example, a trajectory that is a linear approximation between the parameters, and from this new point sequence, a new acoustic It forms a parameter time series Qi(m).

Here, the signal processing in this interpolation point extractor (24) will be explained along the flowchart shown in FIG. First, in block (24a), the value 1 is set to the base J indicating the number in the time series direction of the resampling point, and the value l is set to the variable IC indicating the number in the time series direction of the acoustic parameter time series Pi (nl). is set. Then, the block (
24b), the variable J is incremented, and the block (
In 24c), depending on whether the variable J at that time is (M-1) or higher, it is determined whether the number of the sampling point in the time series direction at that time is the last number that needs to be sampled. If it is, the signal processing of this interpolation point extractor (24) is finished, and if it is not, a block (24d) extracts the signal from the first glue sampling point to the third glue sampling point. The sample distance mDL is calculated, a variable 1G is incremented in block (24e), and block (24f)
The resampling distance DL is the acoustic parameter time series Pi (
Distance 111t S LGc from the first parameter P H1+ of nl to the first C-th parameter PiQc)
), depending on whether the glue sampling point is smaller than the parameter P 1 on the trajectory.
It is determined whether the position is closer to the starting end of the trajectory than (IC), and if it is not located, the variable IC is incremented at block (24e), and then block (24f) is executed again.
The positions of the resampling point and the parameter Pi (Ic) on the trajectory are compared at ＞New acoustic parameter Qi along the trajectory by Niburi sampling
(J) is formed. That is, first, from the resample distance DL at the fifth glue sampling point, the (1(, -1
) Distance 1i1 by parameter P 1 (IC-1) of number H
11sL(Ic-x) is subtracted to find the distance SS from the parameter PiQc-x) of No. 3 H (IC-1) to the sampling point No. 3 No. 1. Next, the parameter P i (Ic-1) and the parameter Pi (Ic) located on both sides of this fifth glue sampling point in the trajectory "2"
distance between %li S 0c-1) (this distance 1jit S
(Ic-1> is obtained by signal processing divided by j in equation (7).) Divide this distance SS by sS/S (I
C-1) L, this division result SS/Sθ (knee 1) is 1
Parameters P iQc ) and P 1 (Ic
-t) (Pi(IC) - Pi(Ic -t)
) multiplied by (Pi(lc) - Pi(Ic-1)) *
55/S(+c-t), and the (IC-1)th parameter Pi(lc-1) located adjacent to the starting end side of the fifth resampling point on the trajectory. Calculate the amount of interpolation from ゛ζ
The located (IC-1)th parameter Pi(+c-
z) to form a new acoustic parameter mouth 1 (J) along the trajectory. Figure 14 shows a two-dimensional acoustic parameter time series P(11, P(21,..., P(81), for which a trajectory is estimated by linear approximation between the parameters, and linear interpolation is performed along this trajectory. 6 new acoustic parameter time series Qfll, Q(2), ..., Q(6
An example of forming 1 is shown below.

Further, in this block (24g), one-dimensional (i=], . . . , I) signal processing is performed in the frequency sequence direction.

In this way, blocks (24b) to (24g) are set at the starting and ending points (Qi(11-pi(ol,
Qi(M)-Pi(S). ) except < (M-
2) A new acoustic parameter time series Qi (@) is formed by point sampling.

This new acoustic parameter time series Qit of the NAT processing unit (21) is stored in the standard pattern memo (January 4) for each recognition target in the registration mode by the mode changeover switch (3), and in the recognition mode. It is supplied to one end of the Chebyshev distance calculating section (25). Also, in this recognition mode, the standard pattern stored in the standard pattern memo i month 4) is supplied to the other end of the Chebyshev distance calculation section (25). In this Chebyshev distance calculation unit (25), a new acoustic parameter time series O1 (input pattern consisting of @ and standard pattern memo January 4) which is normalized on the time axis of the audio input at that time is used. Chebyshev distance calculation processing with the pattern is performed.

Then, the distance signal indicating this Chebyshev distance is supplied to the minimum distance determining section (6), and the minimum distance determining section (6) determines the standard pattern that has the minimum Chebyshev distance with respect to the input pattern, and based on this determination result. A recognition result indicating the input speech is supplied to an output terminal (7).

The operation of the speech recognition device constructed in this way will be explained.

The audio signal of the microphone (1) is converted into an acoustic parameter time series P i (nl), which is a normalized vocal cord sound source characteristic of vocal cord sound source characteristics, for each voice section in the acoustic analysis unit (2), and this acoustic parameter time series Pilnl is subjected to NAT processing. The NAT processing unit (21) estimates a trajectory by linear approximation in the parameter space from the acoustic parameter time series Pi (nl), performs linear interpolation along this trajectory, and performs time axis normalization. A new acoustic parameter time series (old fm) is formed, and in the registration mode, this new acoustic parameter time series Qi(m) is selected by the mode changeover switch (
3) and stored in the standard pattern memory (4).

In the recognition mode, the new acoustic parameter time series old (@) of the NAT processing unit (21) is supplied to the Chebyshev distance calculation unit (25) via the mode changeover switch (3), and is also stored in the standard pattern memory (4). ) is supplied to the Chebyshev distance calculation unit (25).
Parameter time series of one-dimensional input pattern A shown in FIGS. 4 to 6 in FIGS. 5 to 17; 2.4.6.8.
8.8.8.6.4.4.4.6.8, Parameter time series of standard pattern A'; 3.5.7.9°9, 9.
9.7.5.5.7.9, parameter time series collection 7.6.6.8.8.8.8.6.4.4°4 of standard pattern B' to NAT processing unit (21) Parameter time series of one-dimensional input pattern A whose trajectory was estimated by linear approximation and the resampling points were set to 8; 2.4.6.8.
6.4.6.8, Noramameter time series i of standard pattern A' 3.5. "7.9.7.5.7.9, Parameter time series of standard pattern B', 7.6.7.8°7,6
．． 5.4 are shown respectively. In this case, a trajectory in the parameter space is estimated from the acoustic parameter time series Pi(nl, and a new acoustic parameter time series Qi is created along this trajectory.
l(6) is formed, the time axis normalization is performed by the acoustic parameter time series PL (nl itself) obtained by converting the input speech.Then, the Chebyshev distance 1Iltt calculation unit (2
In 5), the Chebyshev distance N8 between the input pattern A and the standard pattern A' is calculated, and the Chebyshev distance 16 between the input pattern A and the standard pattern B' is calculated, and these distances 8 and 16 are calculated, respectively. The distance signal shown in FIG. As a result of the determination, a recognition result indicating that the input voice is the standard pattern A is obtained at the output terminal (7). Therefore, it is possible to perform speech recognition with relatively few erroneous recognitions, even if the speech is partially similar.

Here, the difference in the amount of calculation between a voice recognition device that performs NAT processing and a voice recognition device that performs DP matching processing will be explained.

D of ゛ per standard pattern for input pattern
The average amount of calculation in the P matching distance calculation unit (5) is α
When the average amount of calculation in the Chebyshev distance calculation unit (25) is β and the average amount of calculation in the NAT processing unit (21) is γ, the amount of calculation C1 due to the DP matching process for the five standard patterns is C1− α・J ・ ・ ・ (1
1). Further, the amount of calculation C2 when performing NAT processing on five standard patterns is C2-β·J1γ (12). In general, the average amount of calculations α is α)β compared to the average amount of calculations β
There is a relationship. Therefore, the relationship γ ・ ・ ・ (13) α−β holds true, that is, as the number of words to be recognized increases, the amount of calculation C1 becomes the relationship C1>>C2 with respect to the amount of calculation C2, and the NAT process is Depending on the speech recognition device that performs this, the amount of calculation can be significantly reduced.

In addition, the storage area of the new acoustic parameter time series Qi obtained from the NAT processing unit (21) ((b) can be set to a constant number of parameters in the time series direction, so the storage area of the standard pattern memo January 4) can be effectively used. , and its storage capacity can be relatively small.

By the way, in a speech recognition device that performs such NAT processing, there are standard errors that should not be judged for input pattern A in the situation shown in FIG. 1
B' is taken as the determination result.

In this FIG. 18, input pattern A; "A", standard pattern Δ'; Neo. In this case, input pattern A
contains the same sound M″A″ for the standard pattern B′,
A standard pattern B' in which the input pattern A is more similar to the standard pattern B' than the standard pattern A' in the quasi-stationary region without silence and "A", and the overall trajectory is different but no resembling point should be determined. is approaching.

At this time, the Chebyshev distance of the standard pattern B' with respect to the input pattern A is obtained as a smaller value than the Chebyshev distance of the standard pattern A' in the Chebyshev distance calculation unit (25), and the standard pattern B', which should not be judged, is the judgment result. It is done. In a speech recognition device that performs NAT processing in this way, as shown in FIG. 18, the same phoneme is included, but the overall trajectory is different, and the sampling point may approach the standard pattern B' that should not be determined. , at this time, there was an inconvenience that erroneous recognition occurred and the rate decreased to VAΔ.

Purpose of the Invention In view of the above, an object of the present invention is to obtain a pattern that is relatively less likely to be misrecognized when it approaches a standard pattern that contains the same phoneme, has a different overall trajectory, and whose sampling point should not be determined. Tozuru.

Summary of the Invention The present invention has an audio signal input section, supplies an audio signal from the audio signal input section to an acoustic analysis section, and supplies an acoustic parameter series obtained based on the acoustic analysis section to a trajectory length calculator. ,
This trajectory length calculator calculates the trajectory length of the trajectory in the parameter space from the acoustic parameter series, and the processing result of matching the 12 input patterns and the standard pattern is judged according to the trajectory length of the input pattern and the standard pattern. , which is designed to recognize speech, and according to the speech recognition device of the present invention, it is a standard in which the sampling points should not be determined because they contain the same phoneme but the overall trajectory is different,
There is an advantage that erroneous recognition when approaching a pattern can be relatively reduced.

Embodiment Hereinafter, an embodiment of the speech recognition apparatus of the present invention will be described with reference to FIG. In this figure 19,
The same reference numerals are given to the parts corresponding to those in the figures to FIG. 18, and detailed explanation thereof will be omitted.

In this example, as shown in FIG.
) of the interpolation point extractor (24) is supplied to one end of the trajectory length signal adder (26) and the trajectory length calculator (22) of the NAT processing section (21).
) is supplied to the other end of a trajectory length signal adder (26) and one end of a distance signal corrector (27) to be described later.

This trajectory length signal adder (26) is connected to the NAT processing unit (21)
The trajectory length SL of the trajectory in the parameter space of the acoustic parameter time series Pi(n) of the acoustic analysis unit (2), which becomes a new acoustic parameter time series Qi (@ redundant) !IL trace length double length signal is added.

A new acoustic parameter time series (ifm) to which the trajectory length signal of the trajectory length signal adder (26) has been added is added to the standard pattern memo for each recognition target word in the registration mode by using the mode changeover switch (3). ), and in the recognition mode, it is supplied to one end of the Chebyshev distance calculation unit (25). Further, in this recognition mode, the standard pattern stored in the standard pattern memory (4) is supplied to the other end of the Chebyshev distance calculating section (25). In this Chebyshev distance 1ilIIW output section (25), a signal is formed by adding a trajectory length signal of a standard pattern corresponding to this Chebyshev distance to a distance signal indicating the Chebyshev distance.

The distance signal to which the trajectory length signal of the Chebyshev distance calculator (25) has been added is supplied to the other end of the distance signal corrector (27λ). The UL trace length signal added to the new acoustic parameter time series Q i (ml is compared with the trace length signal of the standard pattern corresponding to the distance signal, and the distance signal is corrected based on the comparison result.

Here, this distance signal compensator (27) will be further explained. In general, if the words are the same, the acoustic parameter series is considered to draw a trajectory in the parameter space with approximately the same shape and length.

Focusing on this point, the distance signal corrector (27)
The distance between the input pattern and the standard pattern (in this example, it is the Chebyshev distance) is corrected according to the difference in trajectory length between the input pattern and the standard pattern. That is, the locus length of the standard pattern is TI? LS, and the trajectory length of the input pattern is TR3I, and the trajectory length T of these standard patterns is
RLS and input pattern trajectory length Tl? The deviation TRL of the UL trace length from LI is calculated by, for example, signal processing. In this case, as is clear from equation (14), the trajectory length deviation TRL takes the maximum value of -++11t2 when the trajectory length TRLS of the standard pattern and the trajectory length TRLI of the input back cover are equal TRLS=TRLI. Then, when the distance signal is Chbs, correction is performed on this distance signal Chbs by signal processing as expressed by the following formula using the trajectory length deviation TRL, so as to obtain a corrected distance signal CHBS. .

CHBS=Chbs-TRLa(a > 0)...(
15) In this example, set a=2.

Corrected distance signal CH of this distance signal supplement (27)
The BS is supplied to the minimum distance determining section (6). Other components such as the acoustic analysis section (2) are constructed in the same manner as the speech recognition device shown in FIG. 9 above.

The operation of the speech □ recognition device constructed in this way will be explained.

The audio signal from the microphone (1) is sent to the acoustic analysis section (2).
ζ The normalized acoustic parameter time series P i (nl) of the vocal cord sound source characteristics is supplied to the NAT processing unit (21) for each speech interval, and the NAT processing unit (21) converts the acoustic parameter time series P i (n) into A trajectory is estimated by linear approximation in the parameter space, and a new acoustic parameter time series Qi (→
is formed. Then, the trajectory length signal adder (26) calculates the trajectory length of the trajectory by linear approximation in the parameter space of the acoustic parameter time series Pi(nl) of the acoustic analyzer (2), which is the source of this new acoustic parameter time series Qi1ml. A trajectory length signal shown is added.

Then, in the registration mode, a new acoustic parameter time series (IH (2)) to which the trajectory length signal of the trajectory length signal adder (26) has been added is transferred to the standard pattern memory (4) via the mode changeover switch (3). ).

In addition, in the recognition mode, the trajectory length signal adder (26)
The new acoustic parameter time series Qi (@ is supplied as an input pattern to the Chebyshev distance calculator (25) via the mode changeover switch (3), and the standard pattern of the standard pattern memo January 4 is input to the Chebyshev distance calculator (25). 2
5) and this Chebyshev distance calculator (25)
The Chebyshev distance between the input pattern and the standard pattern is calculated in Ch.
A signal obtained by adding a standard pattern trajectory length signal corresponding to this Chebyshev distance to bs is supplied to a distance signal corrector (27).

On the other hand, a new acoustic parameter time series Ql(
The trajectory length signal added to (e) is sent to the distance signal corrector (27).
The distance signal corrector (27) calculates the trajectory length TRLI of the input pattern and the trajectory length TRLS of the standard pattern.
The deviation TRL from the trajectory length is obtained by the signal processing shown in equation (14), and the signal processing shown in equation (15) is performed using this trajectory length deviation TRL, and the correction is made based on the trajectory length deviation TRL. A distance signal CHBS is obtained. In this case, as shown in FIG. 18, the standard pattern B', which indicates a different word from the input pattern A, contains the same phoneme "A" with respect to the input pattern, and although the overall trajectory is different, the resampling point approaches, and the Chebyshev Even when the distance is the minimum compared to the standard pattern A' etc. for the same word,
The trajectory length deviation TRL of the standard pattern A' indicating the same word with respect to the input pattern A is approximately equal to the minimum value 2, whereas the trajectory length deviation TRL of the standard pattern B' indicating a different word with respect to the input pattern A is approximately equal to the minimum value 2. Takes a relatively large value. Therefore, the distance signal corrector (27) obtains a corrected distance signal CHBS consisting of the standard pattern A' showing the same sit as the input pattern A, and this corrected distance signal CHBS is transmitted to the minimum distance determining section (6). ), and the standard pattern A' to be determined for the input pattern A is obtained at the output terminal (7) as a determination result.

As described above, according to the speech recognition device of this example, the microphone (11) is provided as an audio signal input section, and the audio signal of this audio signal input section (11) is input into the acoustic parameter time series Pi.
(nl is supplied to the trajectory length calculator (22), and the trajectory length calculator (22) calculates the trajectory length of the trajectory in the parameter space from the acoustic parameter time series Pi(nl), and calculates the trajectory length of the trajectory in the parameter space from the input pattern Since speech is recognized by determining the matching processing result according to the trajectory length of the input pattern and the standard pattern, the sampling point should not be determined because the overall trajectory is different even though it includes the same phoneme. There is an advantage that erroneous recognition can be relatively reduced when approaching a standard pattern.

In the above embodiment, a case was described in which the distance signal corrector (27) performs signal processing expressed by equations (14) and (15), but these (
It is possible to perform signal processing expressed by an appropriate function, not limited to equations (14) and (15). Furthermore, in the above embodiment, the case was described in which the trajectory length of the trajectory in the parameter space was calculated from the acoustic parameter time series Pi(n), but it is also possible to calculate the trajectory length of the trajectory in the parameter space from the acoustic parameter frequency series. It is easy to understand that the same effects as those of the above-mentioned embodiments can be obtained. Furthermore, in the above embodiment, the trajectory length of the trajectory was calculated from the acoustic parameter time series by linear approximation in the parameter space. However, it is easy to understand that the same effects as in the above embodiment can be obtained.

Furthermore, in the above embodiment, the acoustic parameter time series Pi(n) of the acoustic analysis section (2) is
The trajectory length calculator (22) of this NAT processing 1'H4 (21) calculates the trajectory length of the trajectory in the parameter space from the acoustic parameter time series Pi(n). As described above, a trajectory length calculator is provided separately from the 1lill trace length calculator (22) of the NAT processing unit (21), and this!
The new acoustic parameter time series i (ml) of the NAT processing unit (21) is supplied to the FIL trace length calculator, and the locus of the trajectory from the new acoustic parameter time series Qt (2) to that parameter space is calculated. It is easy to understand that the same effect as in the above embodiment can be obtained by calculating the trajectory length and correcting the distance and signal based on this trajectory length.Furthermore, as shown in FIG. Even in a speech recognition device that performs DP ma-noting processing, the acoustic analysis section (
The acoustic parameter series of 2) is supplied to a trajectory length calculator, the trajectory length signal of this trajectory length calculator is added to the acoustic parameter series, and the DP Manning distance is corrected according to the trajectory length of the input pattern and the standard pattern. Even so, misrecognition can be relatively reduced. It goes without saying that the present invention is not limited to the above-described embodiments, and can take various other configurations without departing from the gist of the present invention.

Effects of the Invention According to the speech recognition device of the present invention, it has an audio signal input section, the audio signal from this audio signal input section is used as an anchor to the acoustic analysis section, and the acoustic parameter series obtained based on the acoustic analysis section is obtained. The trajectory length calculator calculates the trajectory length of the trajectory in the parameter space from the acoustic parameter series, and matches the input pattern and the standard pattern. Since the speech is recognized by making a judgment based on the trajectory length, it is relatively possible to make a false recognition when approaching a standard pattern that contains the same phoneme and has a different overall trajectory, but the resampling point should not be judged. There are benefits that can be reduced.

[Brief explanation of the drawing]

Fig. 1 is a configuration diagram showing an example of a speech recognition device that performs speech recognition by DP matching processing, and Fig.
A conceptual diagram for explaining the matching process, FIG. 3 is a diagram for explaining the locus in the acoustic parameter space, and FIGS. 4, 5, and 6 are one-dimensional input pattern A, standard pattern A', and A diagram showing an example of standard pattern B',
Figure 7 is a diagram for explaining time axis normalization by DP matching processing between the parameter time series of input pattern A and the parameter time series of standard pattern A', and Figure 8 is a diagram showing the parameter time series of input pattern A and the standard pattern A'. A diagram for explaining time axis normalization by DP matching processing with the parameter time series of pattern B', FIG. 9 is a configuration diagram showing an example of a speech recognition device that performs speech recognition by performing NAT processing, Figures 10, 11, 12 and 14 are N
Figure 13 is a diagram for explaining the AT processing section, Figure 13 is a flowchart for explaining the interpolation point extractor, Figures 15, 16, and 1.
Figure 7 is a diagram showing the one-dimensional acoustic parameter time series of input pattern A, standard pattern A', and standard pattern B' that were each subjected to NAT processing by the NAT processing unit, and Figure 18 is a diagram showing the entire trajectory including the same phoneme. The route map, the first
FIG. 9 is a block diagram showing an embodiment of the speech recognition device of the present invention. (1) is a microphone as an audio signal input section; (2)
is the acoustic analysis section, (3) is the mode changeover switch, (4) is the standard pattern memory, (6) is the minimum distance determination section, (11) is the
Eight). (11B), ..., (llo) is a 15-channel digital bandpass filter bank, (16
) is the voice interval parameter memory, (21) is the NAT processing unit, (22) is the trajectory length calculator, (23) is the interpolation interval calculator, (24) is the interpolation point extractor, and (25) is Choi Byshiev. A distance 1111t calculating section, (26) a trajectory length signal adder, and (27) a distance signal corrector. Fig. 1 Fig. 13 Fig. 14 Fig. 18 Fig. 19 - Totsukune City - tE'7F)';' io month 11, 1978 Patent Application No. 130714 2°';a'
91ov t+ +lh Iyu4. (Engineering) 43. Relationship with the case of the person making the amendment. Address of the watch applicant: 6-chome, 17th &35', Kitashinyo, Tokyo Parts Store
; Name (218) Sony Co., Ltd. Representative Director Nori Ohga 1F 4, Agent 6, Number of inventions increased by amendment 7, Subject of amendment 8"lj Detailed explanation column of the invention in the specification. 8 , Contents of amendment (1) In the specification, page 34, line 34 to line 7 N (N+
1) (N-1) (N = 15) ... (2) N (N-],
) (N=15) ... (3)'' should be corrected as follows. 1 (1+1) (1-1) (1= 15) ... (2) (1= 15)
... (3)'' (2) Same page, lines 14 to 15 are corrected as follows. (3) Same, page 34, line 3 r TR3IJ
Correct to TRLIJ. (4) Same, page 36, line 7, ``Chezykhiev distance'' is corrected to ``Chebyshev distance.''that's all

Claims (1)

[Claims] comprising an audio signal input section, supplies the audio signal of the audio signal input section to an acoustic analysis section, supplies an acoustic parameter series obtained based on the acoustic analysis section to a trajectory length calculator, and supplies the acoustic parameter series obtained based on the acoustic analysis section to a trajectory length calculator; The length of the trajectory in the parameter space is calculated from the acoustic parameter series, the processing result of matching the input pattern and the standard pattern is determined according to the trajectory length of the input pattern and the standard pattern, and the audio is A speech recognition device characterized in that it recognizes speech.