JPS62136700A - 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] The invention will be explained in the following order.

A. Field of industrial application B. Overview of the invention C. Prior art D. Problem to be solved by the invention E. Means for solving the problem F. Effect G. Example G1 Description of acoustic analysis circuit (Fig. 1) 62 hours Explanation of normalization processing (Fig. 1, Fig. 2) Explanation of specific example of G3 bias value assignment (Fig. 1, Fig. 3) Explanation of G4 pattern matching processing (Fig. 1) H Effect of invention A Industry FIELD OF THE INVENTION The present invention relates to an apparatus for performing speech recognition by performing pattern matching between a standard pattern of a word to be recognized, which is created and stored in advance, and an input pattern of a word to be recognized.

B. Summary of the Invention This invention estimates a trajectory in a device that performs speech recognition by pattern matching an input pattern obtained by estimating a trajectory drawn by an acoustic parameter sequence of a recognition target word with its standard pattern. By applying a bias to the distance between adjacent parameters in the time series direction of the acoustic parameter series used in the process, it is possible to remove the influence of fluctuations in the quasi-stationary region, or to improve the characteristics of the quasi-stationary region. This allows for better extraction.

C Conventional technologySpeech is a phenomenon that changes along the time axis, and is a phenomenon that changes along the time axis.
Unique syllables and words are created by vocalizing sounds with ever-changing patterns. Speech recognition is a technology that automatically recognizes words and phrases spoken by humans.
At present, it is extremely difficult to achieve speech recognition that is comparable to the human auditory function. For this reason, most speech recognition methods currently in practical use are performed by pattern matching a standard pattern of words to be recognized and an input pattern under certain conditions of use.

Figure 4 is a diagram for explaining the outline of this speech recognition device, in which the voice input from the microphone (11) is input to the acoustic analysis circuit (
2). This acoustic analysis circuit (2) extracts acoustic parameters representing the characteristics of the input speech pattern. Various acoustic analysis methods can be considered to extract this acoustic parameter, but for example, a bandpass filter and a rectifier circuit are used as one channel, and a plurality of such channels are arranged with different passbands, and this bandpass filter A method is known in which a temporal change in a spectrum pattern is extracted as a group output. In this case, the acoustic parameters are expressed as their time series Pi(nl (i = 1. 2...
・TRI is, for example, the number of channels of a bandpass filter,
n=1, 2...NUN can be expressed as the number of frames used for recognition in the section determined by voice section determination.

Acoustic parameter time series P from this acoustic analysis circuit (2)
i(nl is a mode switching circuit (for example, a switch)
3). The switch of this circuit (3) is terminal A
When it is switched to the registration mode, it is stored in the acoustic parameter time series Pi (nl is the recognition parameter in the standard pattern memo January 4). That is, prior to speech recognition, the speaker's speech pattern is stored as a standard pattern in this memo. Note that during this registration, the number of frames for each registered standard pattern generally differs due to variations in speaking speed and differences in word length.

On the other hand, when this switch (3) is switched to the terminal B side, it is the recognition mode. In this recognition mode, the acoustic parameter time series of the input voice at that time from the acoustic analysis circuit (2) is supplied to the input voice pattern memo (January 5) and temporarily stored. Then, the distance calculation circuit (6) calculates the difference between this input pattern and each of the standard patterns of the plurality of recognition target words read from the standard pattern memo (January 4). The recognition target word with the smallest difference from the pattern is detected by the minimum value determination circuit (7), and the input word power q is determined by this.

In this way, the input speech is recognized by pattern matching processing between the registered standard pattern and the input pattern, but in this case, even if the same word is uttered in the same way, the spectrum pattern will not change in the time axis direction. You must take into consideration that it may shift or expand/contract. In other words, for example, when recognizing the word "hai", if the standard pattern is registered as "hai", the input voice is "
If the word ``hi'' extends along the time axis, the distance will be very different and it will be treated as a completely different word, making it impossible to recognize it correctly. Therefore, in pattern matching of audio 6+OA&, it is necessary to perform time normalization processing to correct this shift in the time axis direction and expansion/contraction, and this time normalization is an important process for improving recognition accuracy. It is.

One method of time normalization is D P (Dynami
cProgramming) There is a method called matching (for example, see Japanese Patent Application Laid-Open No. 50-96104).
.

This DP matching does not prepare a large number of standard patterns that take time axis deviations into account, but instead generates a standard pattern that normalizes a large number of times using a distortion function, and calculates the distance between this and the input pattern. Speech recognition is performed by detecting the one with the minimum value.

By the way, when using this DP matching method, the number of frames of the standard pattern to be registered is undefined, and it is necessary to perform DP matching processing between all registered standard patterns and the input pattern, and as the number of word groups increases, the amount of calculation increases. The disadvantage is that the amount increases dramatically.

Furthermore, since DP matching is a matching method that emphasizes a stationary part (a part of the spectrum that does not change over time), there is a possibility that erroneous recognition may occur between partially similar patterns.

The present applicant would like to first propose a time normalization method that does not cause such drawbacks (for example, Japanese Patent Application No. 106177/1982).
.

That is, the acoustic parameter time series Pi(n) draws a sequence of points when its parameter space is considered. For example, when the recognition target word is rHAIJ and the number of acoustic analysis band-pass filters is 2 and Pi(nl-(Pt P2)), the acoustic parameter time series of the input speech has the following two-dimensional parameter space: Draw a sequence of points as shown in Figure 5.

As is clear from this figure, the point sequence of the non-stationary part of the voice is distributed coarsely, and the quasi-stationary part is densely distributed. In this case, if the audio is completely stationary, the parameters will not change, and in that case, the point sequence will stay at one point in the parameter space, but humans can produce the same sound without any fluctuation in the audio. Therefore, it does not become completely stationary, and as shown in the figure, it becomes a quasi-stationary part and is affected by fluctuations.

From the above, it is considered that most of the deviation in the time axis direction due to variations in speech rate is caused by differences in point sequence density in the quasi-stationary part, and that the influence of the time length of the unsteady part is small. Therefore, if we estimate the trajectory of the human parameter time series Pi (nl), which is drawn as a continuous curve that approximately passes through the entire point sequence as shown in Figure 6, this trajectory will be the same as that of the voice. It can be seen that it is almost unchanged with respect to variations in speaking speed.

Based on this, the applicant proposed the following time axis normalization method. That is, first, the input parameter time series I
'1 (nl) A trajectory drawn as a continuous curve from the starting point Pi (1) to the ending point Pi (N) is estimated. This is done by approximation.The length S of the trajectory is determined from this estimated curve.

Then, resampling is performed at a predetermined length T along this trajectory as indicated by the circle in FIG. For example, when resampling to M points, T=S/(M 1)...based on the length of (1)! lt traces are resampled.

The parameter time series depicting this resampled point sequence is old (ml (i = 1. 2-1. m = 1.
2 ・−・・M), this parameter time series old (
m) has basic information on the trajectory, and is a parameter that is almost invariant to variations in speech rate.

In other words, it is a recognition parameter time series whose time axis has been normalized.

Therefore, this parameter time series old (m) is registered as a standard pattern, and the input pattern is also obtained as this parameter time series old (ml), and the distance between both patterns is calculated using this parameter time series old (m). If the voice recognition is performed by detecting the one with the minimum distance, the voice recognition will always be performed in a state where the deviation in the time axis direction is normalized and removed.

According to this processing method, the recognition parameter time series old (m
) is always M, and in addition, the recognition parameter time series Qi (ml) is time-normalized, so the calculation of the distance between the input pattern and the registered standard pattern is even the simplest calculation of the Dzievisiev distance. Good effects can be expected.

Furthermore, the above method is a time normalization method that places more emphasis on non-stationary parts of speech, and reduces misrecognition between partially similar patterns such as in DP matching processing.

Furthermore, since the speech rate fluctuation information is not included in the normalized parameter time series Qi(m), it is easy to handle the global characteristics of the parameter transition structure arranged in the parameter space, and it is useful for speaker-independent recognition. Various effective methods can also be applied.

In addition, in the following, the above time normalization process is performed using N A
T (Normalization Along Tr
This is called ajectory processing.

D. Problems to be Solved by the Invention Incidentally, even if the above-described NAT processing is performed, the influence of fluctuations in the quasi-stationary region remains.

Conversely, the characteristics of this quasi-stationary part differ depending on the speaker, so
If more features of this quasi-stationary region can be extracted, the recognition rate may be improved.

This invention is an improvement of the NAT processing method that makes it possible to achieve the two seemingly contradictory things mentioned above, namely, to eliminate the influence of the quasi-stationary region as much as possible, and to be able to extract more characteristics of the quasi-stationary region. This is an attempt to provide a proposal.

E. Means for Solving the Problems The present invention includes an acoustic analysis means (2) for obtaining an acoustic parameter sequence of an input audio signal;
), and a bias applying means (92) that applies a bias to each distance obtained by the inter-parameter distance calculation means (91). and normalization parameter generation means (93) that estimates a trajectory drawn in the parameter space by the acoustic parameter series from the acoustic analysis means (2) based on the biased distance between each parameter, and generates a recognition parameter series from this trajectory. (94)
(95) and the standard pattern memo that stores the recognition parameter series of the standard pattern of the recognition target word (January 4)
and distance calculation means (6) for calculating the difference between the input pattern recognition parameter sequence formed based on the acoustic parameter sequence and the standard pattern recognition parameter sequence from the standard pattern memory; and distance calculation means (6). Minimum value determining means (7) is provided for detecting the minimum standard pattern word of the values calculated in and obtaining a recognition output.

F By subtracting a predetermined bias value from the inter-parameter distance of the acoustic parameter series of the action input, the distance interval between the plurality of parameters in the quasi-stationary part can be made zero or minute,
The quasi-stationary region can be regarded as a stationary region with almost no fluctuations.

Furthermore, if a predetermined bias value is added to the inter-parameter distance of the input acoustic parameter series, the quasi-stationary part, where the inter-parameter distance is originally large, becomes a distance greater than the predetermined value, and can be treated in the same way as an unsteady part, that is, a transient part. This makes it possible to extract the features of this quasi-stationary region.

雫-G Embodiment FIG. 1 shows an embodiment of the speech recognition device according to the present invention. This example shows a case where a 16-channel band-pass filter group is used for acoustic analysis, and the parts corresponding to those in FIG. 4 are the same. Attach a sign.

Description of G1 acoustic analysis circuit (2) In other words, in this example, in the acoustic analysis circuit (2), the audio signal from the microphone (1) is transmitted to the amplifier (21).
1) and a band-limiting low-pass filter (212>) to the A/D converter (213).
It is converted into a 12-bit digital audio signal at a sampling frequency of 2.5kHz. This digital audio signal is
A digital band pass filter (2211) for each channel of a 15 channel band pass filter bank (22)
) , (2212) ,..., , (221x
6). This digital bandpass filter (
22h), (2212),..., (2
2h6) is composed of, for example, a Butterworth fourth-order digital filter, and each band from 25011z to 5.5KI (z is divided at equal intervals on the logarithmic axis) becomes the passband of each filter. Then, each digital band pass filter (2211), (22
12).

The output signals of ..., (22hi,) are respectively rectified by circuits (2221), (2222), ...
, (2221G) and these rectifier circuits (2221G).
22□), (2222).

...The outputs of (2221 & ) are respectively digital low-pass filters (2231), (2232),
..., (2231G). These digital low-pass filters (2231), (2232)
, . . . (223ts) is composed of, for example, an FIR low-pass filter with a cutoff frequency of 52.8 Hz.

Each digital low-pass filter (2231), (2232), etc. which is the output of the acoustic analysis circuit (2)
.

The output signal of (22316) is supplied to the sampler (23H) that constitutes the feature extraction circuit (23). In this sampler (231), the digital low-pass filter (2231
) , (2232) ,..., (2231
The output signal of s) is sampled every frame period of 5.12m5ec. Therefore, from this, a sample time series At(nl (i = 1. 2. -16; n is the frame number and n = 1.2, . . . , N) is obtained.

The output from this sampler (231), i.e., the sample time series ^1 (nl), is supplied to the sound source information normalization circuit (232), which removes differences in vocal cord sound source characteristics depending on the speaker of the speech to be recognized. be done.

That is, for the sample time series Ai[n) supplied from the sampler (231) every frame period, Ai(n)=
log (^1(nl+B)...
・(2) A logarithmic transformation is performed. In this equation (11), B is set to a value that hides the noise level with the bias.

Then, when the vocal cord sound source characteristics are approximated by the formula yi=a−i+b, the coefficients of a and b are determined by the following formula.

(I = 16) ... (3) (1 = 16
) ・ ・ ・ (4) Then, if the normalized parameters of the sound source are Pi [nl, then a fn)
< When O, the parameter Pi(n) is Pi(nl=Ai
fnl −(a (nl −i +b(nl) ・
...(Represented as 51.

Also, when afn)≧0, only level normalization is performed,
The parameter Pi(nl is expressed as...(6).

In this way, the sound source information normalization circuit (232) obtains an acoustic parameter time series Pi(n) in which differences in the vocal cord sound source characteristics are normalized and removed.

The acoustic parameters Pi (nl) from this sound source information normalization circuit (232) are supplied to the voice interval parameter memory (8). In response to the speech section determination signal, a parameter Pi(n) is stored for each determined speech section.

The voice section determination circuit (24) includes a zero cross counter (24).
1), a power calculation circuit (242), and a voice section determination circuit (
243), and the digital audio signal from the A/D converter (213) is supplied to the zero cross counter (241) and the power calculation circuit (242). One frame period of zero cross counter (241) is 5.12m5e
c, the number of zero crossings of the 64 samples of the digital audio signal within this one frame period is counted, and the count value is supplied to the first input terminal of the audio section determining circuit (243). The power calculation circuit (242) calculates the power of the digital audio signal within this l frame period for each frame period, that is, the sum of squares, and the output power signal is sent to the second voice section determination circuit (243). Supplied to the input end. The voice section determining circuit (243) further includes a third
The sound source normalization information from the sound source information normalization circuit (232) is supplied to the input terminal of the sound source information normalization circuit (232). In this speech section determination circuit (243>), the number of zero crossings, the power within the section, and the sound source normalization information are processed in a composite manner, and a process for determining silence, unvoiced sound, and voiced sound is performed, and a speech section is determined.

The voice interval determination signal indicating the determined voice interval from the voice interval determination circuit (243) is transmitted to the voice interval determination circuit (243).
) as the output of the voice interval parameter memory (200
') is supplied.

In this way, within the judgment voice section, the memory (200)
The acoustic parameter time series pi(n) stored in is N
A 'r' processing circuit (9) is supplied.

Explanation of G2 time normalization processing In this case, the NAT processing circuit (9) processes the interparameter distance calculation circuit (91), the bias application circuit (92), the trajectory length calculation circuit (93), the interpolation interval calculation circuit (94), and the interpolation It consists of a point extraction circuit (95).

Parameter time series Pi(nl (i = 1. 2. ”, 16;
n=1. 2°..., N) are supplied to an inter-parameter distance calculation circuit (91). In this inter-parameter distance calculation circuit (91), the acoustic parameter time series Pi
(nl calculates the distance between each parameter in the trajectory drawn by linear approximation as shown in FIG. 7 above in the parameter space B).

In this case, the Euclidean distance D(al, b;) between the ■dimensional vectors ai and b1 is 5(nl=D(Pi(n+1), Pi(
n)) (n=1...., N)
・ ・ ・It is expressed as (8).

The inter-parameter distance S (nl) thus calculated is supplied to the bias applying circuit (92).

In this bias applying circuit (92), a predetermined bias value is subtracted or added to each parameter distance S (n) as described later.

This inter-parameter distance S (inter-parameter distance BSTfl with a bias value given to nl) is supplied to the gL trace length calculation circuit (93), and is calculated from the first parameter P in) to the N-th parameter in the time series direction. The total trajectory length SL up to the (last) parameter Pi(tJ) is calculated using this inter-parameter distance BS(n).

That is, the first parameter P in the time series direction
1 (distance 5Lfn from 11 to the n-th parameter Pifnl) is expressed as. Then, the total trajectory length SL is expressed as.

The trajectory length SL obtained by this trajectory length calculation circuit (93)
A signal indicating the interpolation interval calculation circuit (94) is supplied to the interpolation interval calculation circuit (94).

This interpolation interval calculation circuit (94) calculates the resampling interval T when resampling is performed along the locus.

In this case, if resampling is performed at M points, the resampling interval T is T=SL/(M-1) (11
) is required.

A signal indicating the resampling interval T from this interpolation interval calculation circuit (94) is supplied to an interpolation point extraction circuit (95). In addition, the acoustic parameter time series Pi(n) from the parameter memory (200) and the bias applying circuit (92)
The biased interparameter distance BS(nl
is supplied to this interpolation point extraction circuit (95). This interpolation point extraction circuit (95) uses the acoustic parameter time series Pi(n
The locus of l in its parameter space, for example I!, which is a linear approximation between parameters. l! Re-sampling is performed along the lL trace at a re-sampling interval T as indicated by the circle in FIG. 7, and a recognition parameter time series Qi (ml) is formed from the new point sequence obtained by this sampling. In this case, A biased value BS (n=) is used as the distance between two parameters used during interpolation.

That is, in this interpolation point extraction circuit (95), the second
The processing according to the flowchart shown in the figure is performed, and LUA
A six-parameter time series Qi (ml) is formed.

First, in step [101], a value 1 is set to the variable J indicating the number of the resampling point in the time series direction, and a value l is set to the variable IC indicating the frame number of the acoustic parameter time series Pi (nl). Next, in step (102), variable J is incremented, and in step (103), variable J at that time is (M
-1) It is determined whether the number of the resampling point at that time in the time series direction is the last number that needs to be resampled by determining whether or not it is less than or equal to -1). If it is the last number, step [1
04) and the resampling ends.

If it is not the last number, then in step (105) the first resampling point (this is, for example, a silent part) is selected.

) to the third resampling point is calculated.

Next, the process proceeds to step (106), where the variable IC is incremented.

Next, in step (107), the resampling distance DL is set to the first parameter P of the acoustic parameter time series Pi (nl).
i(distance S from 11 to 1Cth parameter Pir+o
It is determined whether the resampling point at that time is located closer to the starting point of the trajectory than the parameter Pioc+ at that time on the trajectory, and if it is not located closer to the starting point, the process returns to step (106) and the variable IC is incremented. In step (107) again, the positions of the resampling point and the parameter Pi (10) on the trajectory are compared, and the resampling point and the parameter Pi (Io) are compared on the trajectory.
When it is determined that the position is closer to the starting point than the step (
Proceeding to step 10B), the recognition parameters Qiσ) are formed.

That is, from the resampling distance #DL by the third resampling point, the (IC-1)th parameter P located closer to the starting point than this third resampling point
1oc-□, the distance S L (Ic-1) is subtracted to obtain the (IC-1)th parameter P i (IC-1+
Find the distance SS from to the third resampling point. This distance is of course determined using the value BS(n) after adding the bias.

Next, the distance between the parameter P i (Ic-u and the parameter Pioo located on both sides of this third resampling point on the trajectory S (nl) and the subsequent distance 1i11tBs(n ) and divide this distance !1iltss by the division result SS/BS
(At +0., the parameters Pino and Pfoe- located on both sides of the third resampling point on the trajectory
11 (P i (Io P ioc −o )), the third resampling point on the trajectory is located adjacent to the starting point side of the third resampling point (IC-1). The interpolation amount from the th parameter P 1uc-11 is calculated, and this interpolation amount and the th (IC-1
)-th parameter P 1oc-1, and a new recognition parameter Qi (technique) along the trajectory is obtained.

is formed.

In this way, the recognition parameter time series Q (m) /! l< is formed.

Description of a specific example of giving a G3 bias value Various ways of giving a bias value can be considered, but the first example is the minimum value of the distance between parameters S fn) mt calculated by the distance between parameters calculation circuit (91). An example of this is when the bias value is subtracted from the distance S (n) between each parameter. This is the case in which the quasi-stationary region can almost be regarded as a stationary region. That is, in the bias applying circuit (92), BS(nl=s(nl Sfn)mtn
HH+ (12
) is performed.

For example, a two-dimensional parameter time series as shown in Figure 3,
When the distance between each parameter S (nl is as shown in the figure), the minimum value s (n)□tn=3.

Therefore, if the bias applying circuit (92) calculates equation (12), the distance between each parameter BSfn) becomes a value as shown in the lower part of the figure, and the distance between parameters BS(n) in the quasi-stationary part becomes It becomes zero or minute.

Then, in the interpolation point extraction circuit (95), the distance to which this bias value is applied is used to extract the interpolation point in step (108) of the flowchart shown in FIG. The assumed trajectory is estimated,
The recognition parameter time series Gifml is obtained by resampling along the trajectory.

In this way, NAT can reduce the influence of fluctuations in the quasi-stationary region.
In the processing, it is possible to obtain a recognition parameter time series old (m) that can more effectively eliminate the influence of fluctuations in the quasi-stationary region.

Next, as a second example, a bias value a is added to the inter-parameter distance S in). That is, in the bias applying circuit (92), BSfn)=5fn)+a...
- The calculation (13) is performed.

In this example, the new parameter distance BS[+1
Since the distance between the parameters of the quasi-stationary part 1 is extended, the recognition parameter time series Qi (m) obtained from the interpolation point extraction circuit (95) also extracts the features of this quasi-stationary part.

In the case of such bias value addition, the concept of matching window in DP matching is used, and when a=+■, N
AT processing is equivalent to linear stretching.

Note that the above bias value may be changed depending on the trajectory length, or furthermore, the bias value may be determined from the average value of distances between parameters in the quasi-stationary part.

In addition, a new parameter pressure 4 BS (nl) can be obtained by subtracting the bias value from the parameter pressure Ati S (n) to almost eliminate the influence of the quasi-stationary region, and the characteristics of the quasi-stationary region can be extracted more easily. In this case, instead of using the minimum value S (n) □1 of the distance S (n) as described above or the control value as described above, use a fixed value determined through experiments etc. as the bias value. It's okay.

Note that when this bias value is subtracted from the distance % Stt S (nl), it is performed within the range of distance BS[n]≧0 after bias subtraction. However, if BS(nl < 0), the distance 'R?I BS(n) = Q may be forcibly determined.

In addition, although the case where bias is given to the value of the parameter consisting of 16 channels has been explained above, 16
Detailed feature extraction can be performed by considering a parameter for each channel or for each of a plurality of channels, that is, for each frequency band, and performing NAT processing that takes biasing into consideration for the parameter.

G. Description of pattern matching processing The recognition parameter time series Qi (m) from this NAT processing circuit (9) is supplied to the mode switching circuit (3), and the calculated gt trace from the trajectory length calculation circuit (91) is A signal indicating the length is supplied to the mode switching circuit (31).

At the time of registration, the recognition parameter time series is stored in the standard pattern memo (January 4).

Next, during speech recognition, pattern matching processing is performed as follows.

That is, in the NAT processing circuit (9), the NA
The recognition parameter time series i (m
) is the distance calculation circuit (6) via the mode switching circuit (3).
is supplied to calculate the distance to the standard pattern.

The distance in this case is calculated as, for example, a simple Chebyshev distance. The calculation output of the distance between each standard pattern and the input pattern from this distance calculation circuit (6) is supplied to the minimum value determination circuit (7), which determines the standard pattern with the minimum distance calculation value, and based on this determination result. The recognition result of the input speech is obtained at the output end (70).

In addition, in the above embodiment, the acoustic parameter time series P
Although the case has been described in which the trajectory length of the trajectory in the parameter space is calculated from i(n), the trajectory length of the trajectory in the parameter space may be calculated from the sound parameter frequency series.

Also, in the above embodiment, linear approximation is used! Although the trajectory length of the It trace is calculated, the trajectory length of the trajectory may be calculated by arc approximation, spline approximation, or the like.

Effects of the Invention H As described above, according to the present invention, a bias is applied when calculating the distance between parameters in NAT processing, so that when this bias value is negative, the stationary part (quasi-stationary part) ) can be obtained by extracting features only from the transient part.On the other hand, when the bias value is positive, there is no extreme time axis normalization for the stationary part, and this quasi-stationary part It becomes possible to extract the features of

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the device of this invention, FIG. 2 is a flowchart for explaining the operation of the main part thereof, and FIG. 3 is a diagram showing the operation of the main part of the invention. 4 is a block diagram showing the basic configuration of the speech recognition device,
FIGS. 5 to 7 are diagrams for explaining NAT processing. (2) is a joint analysis circuit, (4) is a standard pattern memory,
(6) is a distance calculation circuit between the standard pattern and the human cover turn, (7) is a minimum value judgment circuit, (9) is a NAT processing circuit,
(91) is an inter-parameter distance calculation circuit, (92) is a bias applying circuit, and (95) is an interpolation point extraction circuit. Flowchart for extracting points between buckets Figure 2 A diagram of one sequence of operations for adding a space Figure 3 A block diagram of the basic configuration of an audio magazine IR Figure 4 Parameter space 1: E Diagram showing WJs diagram Parameter space turtle; Draw<S trace n example shown in Figure 6 LL line delay 1: J6 Estimated car 7L trace 0 I column ε Figure 7

Claims (1)

[Scope of Claims] (a) Acoustic analysis means for obtaining an acoustic parameter sequence of an input audio signal; (b) A parameter for calculating the distance between adjacent parameters in the chronological direction of the acoustic parameter sequence from this acoustic analysis means. (c) bias applying means for applying a bias to each distance calculated by the inter-parameter distance calculating means; and (d) the acoustic analysis means based on the distance between each parameter to which the bias has been applied. (e) a standard pattern in which a recognition parameter sequence of a standard pattern of a recognition target word is stored; (f) a distance calculation means for calculating a difference between the recognition parameter series of the input pattern and the recognition parameter series of the standard pattern read from the standard pattern memory; (g) a value calculated by the distance calculation means; A speech recognition device comprising minimum value determining means for obtaining a recognition output by detecting the minimum standard pattern of words.