JPS5852694A - Monosyllabic voice recognition system - 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 provides a monosyllabic speech recognition system, particularly a monosyllabic speech recognition system that performs recognition processing by extracting feature parameter time series based on frequency analysis results of speech signals. Divide the input feature parameter time series into a small number of intervals, extract a reduced feature parameter time series consisting of an average value obtained by averaging the parameter values for each interval, and use the reduced feature parameter time series to identify candidate monosyllables for recognition. The present invention relates to a monosyllabic speech recognition system in which the processing speed is greatly improved by selecting candidate monosyllables and comparing the candidate monosyllables.

In a monosyllabic speech recognition system, the frequency analysis results of a monosyllabic speech signal are used to extract feature parameters and meters that represent the characteristics of each phoneme, and the extracted feature parameters and pre-registered meters corresponding to the registered monosyllables are used. The unknown input monosyllabic speech is recognized by comparing it with the feature parameters. That is, for example, the first formant frequency and the second formant frequency are sampled as the characteristic parameters, and all of these parameters are used.

However, as the amount of data processed in the above matching increases and the number of recognized categories increases, the time required for the above matching process increases.

For this reason, we focused on the fact that there are sections where the above characteristic parameters change rapidly over time and sections where they change gently.
A method is being considered that increases the recognition rate with a small number of samples by sampling densely in the former interval and sparsely sampling in the latter interval, that is, by sampling at non-uniform sampling points ( (Japanese Patent Application No. 52-43972).

For this method, K
various methods to improve processing speed (Patent Application No. 53-53)
No. 3965, Patent Application No. 53966, Patent Application No. 1983-
No. 53967) has been proposed, but the extraction algorithm for extracting the feature cover is complicated and it is difficult to narrow down the candidates significantly.The purpose of the present invention is to improve the above points. The overall purpose is to efficiently determine monosyllable candidates for recognition using a relatively simple algorithm.The overall purpose is to improve the monosyllable speech recognition rate and reduce the recognition processing time. Therefore, the monosyllabic speech recognition system of the present invention analyzes an unknown input monosyllabic speech signal and compares the input feature parameter time series extracted from the speech signal with the registered feature parameter time series registered in advance. In a monosyllabic speech recognition system that recognizes an unknown input monosyllabic speech, the input feature variation from the beginning of the unknown input monosyllabic speech to the vowel included in the monosyllabic speech is determined. The evening time series is divided into at most 0 or less intervals, and an input reduced parameter time series is extracted consisting of the averaged value of all parameter values in each interval or interval boundary values,
The method is characterized in that recognition target candidate monosyllables are determined by comparing them with a registered reduction parameter time series that has been extracted using the same method and registered in advance. This will be explained below with reference to the drawings.

FIG. 1 is an explanatory diagram illustrating the concept of an embodiment of the present invention,
FIG. 2 is an explanatory diagram fully explaining the idea of another embodiment of the present invention, and FIG. 3 shows the configuration of an embodiment of the present invention that performs the above processing.
FIG. 4 shows an explanatory diagram in the form of a flowchart of a process for determining a section in an embodiment corresponding to FIG. 2 above.

As shown in FIG. 1, when it is assumed that the sampled feature parameter P exists between time To and T, t, in the case of the first embodiment of the present invention, from time To to Tr
For example, the time at t is divided into five equal points Tg15.2T
E15, 3 TE01, 4. Ti2B, Tz
Determine.

Then, each feature parameter value between time To and Tz15t is averaged, and each feature parameter value between time TE15 and 2TK15 is averaged, and...
. . . Each feature parameter value between time 4Tg157i and T is averaged, and a reduced parameter time series of, for example, five average value parameters is extracted.

Note that instead of averaging the feature parameter values, a reduced parameter time series consisting of interval boundary values may be extracted.

In extracting the above-mentioned reduced parameter time series, for monosyllabic speech, for example, the sound of "ma (mα)", the starting point To
should be taken at the beginning of the monosyllabic sound, that is, at the beginning of the 1m sound. It is conceivable to set the final point Tg at the final end of a monosyllabic voice, but as shown in the "α" section in Figure 1, the vowel has a part that shows relatively stable constancy, i.e. There is a part where the characteristic parameter value changes little, and it is desirable to set this point as Tg as the vowel representative point. By doing so, it is possible to eliminate the rear portion of the vowel portion of monosyllabic speech that includes unstable elements, and it is possible to improve the recognition rate.

In the case of the second embodiment of the present invention, in contrast to the above-described first embodiment, in which sections are divided at equal intervals on the time axis, the interval between sections is reduced at points where the rate of change of the characteristic parameter is relatively large. I try to choose. That is, when the characteristic parameter P is assumed to be as shown in FIG. 1, the cumulative amount of variation of this parameter P, that is, the cumulative amount of variation, is plotted by plotting time on the horizontal axis (as shown in FIG. 2). For a figure drawn as follows, the cumulative fluctuation cover is the above value −TAV, −gTAV
,......The point in time when TI , Tg ,...
・・・・・・・・・・・・Tg”fir extraction, time To
The values of each feature parameter shown in FIG. 1 from time T1 to Tlt are averaged, and the values of each feature parameter shown in FIG. 1 from time T1 to Tg are averaged. ...,
The characteristic parameter values shown in FIG. 1 from time T4 to Tz are averaged, and a reduced parameter time series consisting of, for example, five average value parameters is extracted.

Considering the average value parameter of the above-mentioned reduced parameter time series, the average value parameter can be considered as follows. For example, as the characteristic parameters P shown in FIG. 1 used in the present invention, parameters M, 2 (1, n, , ) and
tL), each parameter is expressed as follows, as is well known.

(However, No. 1.2) The characteristic parameter MJ (trL) shown in equation (1) is a parameter based on the method of moments, and the parameter M
l(ttL) corresponds to the 17th formant, and parameter Mz(trL) corresponds to the 2nd formant. Further, the characteristic parameter XJ (trL) shown in equation (2) is a parameter corresponding to the power for each band, and the parameter
l('tL) corresponds to the normalized low-frequency power, and the parameter
2(ttL) corresponds to normalized high frequency power.

Assuming that the current section is divided into speed values, the above average value parameter corresponds to the parameter 'xJ'('4L) of equation (1) and the parameter 'xJ' (!rL) of equation (2). Each is expressed as follows. That is, t = -・Tz... and curved surface... (5) Lk
Nf Also, time TI in the case described with reference to FIG.

T2+...... is expressed as follows.

here, (However, V (in): amount of variation) Also, the amount of variation V (trL) in the second part (7) is 9- (However, P is the output of the bandpass filter group) It is given by nine.

The interval determined by the cumulative amount of fluctuation as described above may be determined independently for each parameter, such as a parameter based on the method of moments or a parameter corresponding to band-specific power.

FIG. 3 shows the configuration of an embodiment of the present invention. In the figure, 1 is a band filter group, 2 is a parameter extraction circuit, 3 is a vowel representative point determination circuit, 4 is an input feature parameter time series buffer, 5 is a parameter average interval determination circuit, 6 is a parameter average circuit, 7 and 8 9 are switching circuits that switch between the registration mode and the recognition mode; 9 is a registered monosyllable reduction parameter time series registration unit which is constituted by memory; and 10 is a reduction parameter time series matching unit/candidate determination unit. , 11 is a registered monosyllable feature parameter time series 10-sequence registration unit which is constituted by a memory that stores feature parameter time series for each registered monosyllable;
Reference numeral 2 denotes a candidate selection circuit which selects only those corresponding to recognition target candidate monosyllables according to the present invention from among the feature parameter time series read out from the registration unit 11; 13 a feature parameter time series/verification circuit; The determination unit 14 represents an output circuit.

An input monosyllabic speech signal is input to a bandpass filter group 1, and a parameter extraction circuit 2 extracts input feature parameters corresponding to the input monosyllabic speech signal. The extracted input feature parameters are input to the vowel representative point determination circuit 3, and the vowel representative point determination circuit 3 determines the vowel representative point to use the vowel representative point as the time TE, as explained using FIG. Detect the part that shows stationarity. The input feature parameters up to the vowel representative point determined here are temporarily set in the buffer 4 in the form of input feature parameter time series. The parameter average interval determination circuit 5 operates at the time shown in FIG. -TE
:ATE, In the case of ........., we drew the point Tg at 5'5, and divided the period from To to TE into 5 equal parts at time 2, 100 Tg, 100 TE. , . . . , TE is determined. Time points shown in Figure 2 T 11 T 2...
... will be described later with reference to FIG. 1. Once the sections are determined based on the time point 1, the parameter averaging circuit 6 calculates the average value of the parameter values for each section based on the contents of the buffer 4.

In the registration mode, the switching circuits 7 and 8 follow the route shown in the upper part of the figure. Then, the reduced parameter time series (in this case, the registered monosyllabic reduced parameter time series) extracted by the parameter averaging circuit 6 are registered in the graphical registration section 9, and the feature parameter time series set in the graphical buffer 4 are registered in graphical registration. It is registered in Section 11.

In the recognition mode, the switching circuits 7 and 8 take the lower route shown in the figure. And the parameter average mouth #! The reduced parameter time series (in this case, the input reduced parameter time series) extracted by r6 is led to the graphic matching unit/candidate determining unit 10. At this time, the registered monosyllable reduction parameter time series is sequentially outputted from the illustrated registration unit 9, and is compared with the input reduction parameter time series to determine all the recognition target candidate monosyllables according to the present invention. calculates, for example, the inter-monosyllabic distance Sr, that is, the Chebyshev distance between the registered reduction parameter time series and the manual reduction parameter time series. The distance Sr is given by the following equation and corresponds to the input reduction parameter time series.

When the matching section/candidate determining section 10 determines several candidate monosyllables to be recognized based on the above equation (9), the candidate selecting circuit 12 is notified of the names of the candidate monosyllables. At this time, the feature parameter time series are sequentially read out from the illustrated registration section 11 under control from a control section (not shown). Then, the candidate selection circuit 12 selects only the registered monosyllable feature parameter time series corresponding to the monosyllable designated as a candidate monosyllable, and guides the selected monosyllable to the illustrated comparison determination unit 13. In the case of the above-mentioned recognition mode, the switching circuit 7 selects the router at the bottom of the diagram, and the feature parameter time series (at this time, the input feature parameter time series) set in the diagram buffer 4 is guided to the matching determination section 13. . Thereby, the input feature parameter time series is compared with each registered monosyllable feature parameter time series corresponding to the candidate monosyllable. In this case, (well-known dynamic method)
It may be assumed that programming (DP) verification is performed. In this way, one single syllable category extracted in (7) is output to the output circuit 14.

Time TI+T2+ as shown in the above figure 2
When determining . That is, (1) Based on the parameters extracted by the harameter extraction circuit 2, the cumulative variation ii: TAVi as shown in FIG. 2 is extracted independently for each parameter.

(2) Then, a value DTAV is determined by dividing the cumulative variation amount TAV into five equal parts, for example.

(3) First, set J = 1 to find the time Tl, set the above value DTAV in the register AVH, and set the value T (I) 'e in the timing start register TS(J).
Set above.

(4) Thereafter, the feature parameter values are sequentially accumulated until the cumulative value AV(I) of the feature parameters is equal to the contents of the register AVH, but is still a dog.

(5) When the cumulative value AV(I) is equal to or greater than the contents of the register AVI(), the timing value T(
I) is set in the register TE(I) for time Tl, the value T(I+1) is set in the register TS (J+1), the value (AVH+DTAV) is set in the register AVH, and J is set to be determined at the next time T2. =2.

(6) Similarly, the feature parameter values are accumulated until the cumulative value AV(I) is equal to or equal to the contents of the register AVH. That is, time T21T3. Determining T4fc 0 (7) Then, when the accumulation circuit ■ reaches the value N, that is, when the accumulation process reaches the accumulation of the characteristic parameters corresponding to the time TK shown in FIG. 2, the time Tz is determined at that time. Ru.

As explained above, according to the present invention, recognition target monosyllable candidates can be narrowed down relatively easily and efficiently.
It is possible to significantly reduce the recognition processing time.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram illustrating the concept of an embodiment of the present invention,
FIG. 2 is an explanatory diagram illustrating the concept of another embodiment of the present invention, and FIG. 3 shows the configuration of an embodiment of the present invention that performs the above processing.
FIG. 4 shows an explanatory diagram in the form of a flowchart of a process for determining a section in an embodiment corresponding to FIG. 2 above. In the figure, P is a feature parameter, 2 is a parameter extraction circuit,
3 is a vowel representative point determination circuit, 4 is an input feature parameter time series buffer, 5 is a parameter average interval determination circuit, 6 is a parameter average circuit, 7 and 8 are respectively switching paths, 9 is a registered monosyllable reduction parameter time series registration 10 is an mφ parameter time series matching unit/candidate determining unit, 11 is a registered monosyllabic feature parameter time series registration unit, 12 is a candidate selection circuit, 13
represents the feature parameter time series/matching determination unit. Patent applicant Fujitsu Ltd. agent Patent attorney Mori 1) Kan 11th

Claims (1)

[Claims] A unit that analyzes an unknown input monosyllabic speech signal and compares the input feature parameter time series extracted from the speech signal with the registered feature parameter time series registered in advance to recognize the unknown input monosyllabic speech. In the syllabic speech recognition system, the input feature parameter time series from the beginning of the unknown input monosyllabic speech to the vowel included in the monosyllabic speech is divided into at most 10 sections, and the parameters within each section are It is configured to extract an input reduced parameter time series consisting of average values or interval boundary values, and to identify recognition target candidate monosyllables by comparing it with a registered reduced parameter time series extracted using the same method and registered in advance. A monosyllabic speech recognition system characterized in that it determines.