JPS61183697A - Monosyllable 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 Field of the Invention The present invention relates to a monosyllabic speech recognition device that recognizes input monosyllabic speech using a pre-registered monosyllabic speech standard.

Conventional technology Conventional monosyllabic speech recognition devices include, for example, the Acoustical Society of Japan Speech Study Group Material 1 Development of a Small Monosyllabic Recognition Device” 5e.
a-41, 1983-10.

FIG. 3 shows the configuration of this conventional monosyllabic speech recognition device, in which 1 is an input section for converting speech into a digitalized electrical signal, and 2 is an input section for converting speech into a digitalized electrical signal. 3 is a feature extraction unit that extracts voice features from a signal by means such as frequency analysis; 3 is a vowel-consonant boundary detection unit that detects the boundary between a vowel part and a consonant part of a monosyllabic input pattern and separates the vowel part and the consonant part; Section 4 is a vowel input pattern storage section for storing separated vowel patterns; 5 is a consonant input pattern storage section for storing separated consonant patterns; and 6 is for storing several vowel patterns in advance. 7 is a consonant standard pattern storage section that stores some consonant patterns in advance; 8 is a storage section that stores the patterns stored in the vowel input pattern storage section 4 and the vowel standard pattern storage section; A vowel similarity calculation unit 9 calculates the similarity between the input pattern and the consonant similarity calculation unit 9, which is stored in the consonant input pattern storage unit 6. This is an output unit that determines whether the syllable is recognized and outputs it.

Regarding the monosyllabic speech recognition device configured as above,
The operation will be specifically explained below.

The uttered monosyllabic speech is input to the input section 1, which converts this speech signal into an electrical signal and further digitizes it. This signal is frequency-analyzed using a digital filter for feature extraction. The speech pattern whose features have been extracted in this way is separated into a vowel part and a consonant part by the vowel-consonant boundary detection section 3. This situation is shown in Figure 4a.

Various methods of vowel-consonant boundary detection are possible, but here, as an example, a method of applying a trapezoidal template to the power pattern of the voice will be described. That is, as shown in FIG. 4, the distance 13 between the speech power pattern 11 and the template 12 is calculated, and the minimum point is taken as the vowel/consonant boundary. The vowel patterns and consonant patterns thus separated are stored in the vowel input pattern storage section 4 and the consonant input pattern storage section 5, respectively.

When the recognition device is in the registration mode, the input patterns in the vowel input pattern storage section 4 and the consonant input pattern storage section 6 are stored in the vowel standard pattern storage section 6 and the consonant standard pattern storage section 7, respectively.
is stored in

Next, in the recognition mode, the vowel similarity calculation section 8 outputs the number corresponding to the vowel standard pattern closest to the vowel input pattern to the output section 10;
Accordingly, the number corresponding to the consonant standard pattern closest to the consonant input pattern is output to the output unit 10. The output unit 1o integrates the consonant recognition results and the vowel recognition results and outputs a monosyllable recognition result.

Problems to be Solved by the Invention In the monosyllable recognition device having the above configuration, the following problems occur. In other words, there is a risk that the vowel-consonant boundary detection section 3 will make a large error in the boundary position. FIG. 6 shows a case where the vowel-consonant boundary detection section 3 makes a mistake in boundary detection. If a boundary is detected incorrectly in this way, it becomes a major cause of misrecognition for both consonant and vowel parts.

In view of this, an object of the present invention is to provide a monosyllabic speech recognition device that can reduce the detection error of the vowel-consonant boundary position and improve the recognition rate.

Means for Solving the Problems The present invention calculates a cumulative function of changes in feature amounts of monosyllabic speech prior to vowel-consonant boundary detection, and calculates a cumulative function of changes in feature amounts of monosyllabic speech. A cumulative function calculating section is provided which uses the point at which the amount has increased as a temporary vowel-consonant boundary, thereby limiting the vowel-consonant boundary range.

Function: The present invention can limit the existence of vowel-consonant boundaries within the range of - foot by calculating a cumulative function for input monosyllabic speech. Therefore, even if the vowel-consonant boundary detection unit 3 makes an error, the error can be limited to within a certain range, so large deviations that can cause misrecognition can be avoided, and the recognition rate can be improved. Improvements can be made.

Embodiment FIG. 1 shows the configuration of a monosyllabic speech recognition apparatus according to an embodiment of the present invention. In the figure, 1 is an input section, 2
3 is a feature extraction section, 3 is a vowel-consonant boundary detection section, 4-digit vowel input pattern storage section, 6 is a consonant input pattern storage section, and 6 is a vowel standard pattern storage section, which is the same as that shown in FIG. 3. . The difference from the configuration shown in FIG. 3 is that the cumulative function calculation section is provided between the feature extraction section 2 and the vowel-consonant boundary detection section 3.

The operation of the monosyllabic speech recognition device according to the embodiment of the present invention configured as described above will be described below.

The uttered monosyllabic speech is input to the input section 1, which converts this speech signal into an electrical signal and further digitizes it. Features of this signal are extracted in the feature extraction section 2 by means such as wave number analysis. FIG. 2a shows, as an example, the frequency analysis of monosyllabic speech using 16 digital filters. A cumulative function calculation unit 11 calculates a cumulative function of the voice pattern whose features have been extracted in this way.

That is, if monosyllabic speech is converted into a frequency spectrum pattern series averaged over a certain period of time T, and this is denoted as F (n), then F (n) = (!0 (n), !1 (n), · -----・,
! rn(n)) ・−・−(1) Here, n is the frame number when the time interval T is one frame, xi
(n) (i=0.1,..., m) is n
Indicates the frequency components of the frame. Next, the amount of change in Ii between adjacent frames is accumulated, and this is calculated using the accumulation function accf(
n), then if the value of the above cumulative function at the time when monosyllabic speech is finished being captured is accf (N), and the correct vowel-consonant boundary position is the nsth frame, then the following values α, ε are obtained. can be determined.

accf (na)=accf (N) X −, old,
, (3) α ncL−ε< ns < na+ε
(4) Figure 2 shows, as an example, the cumulative function of Figure 2a. Next, the vowel-consonant boundary existence range equation (4) is determined using α and ε of the above equations (3), 9, and (4), which are experimentally determined in advance. Figure 2C
For example, α=2. from the cumulative function in Figure 2. ε=1
6 shows how the range of existence of the vowel-consonant boundary is determined. Next, information on this range of existence is determined by the vowel −
It is sent to the consonant boundary detection unit 3, where a vowel-consonant boundary is determined between the above existing ranges, and the vowel part and the consonant part are separated.

This situation is shown in FIG. 2d. Here, a method of applying a trapezoidal template to the audio power pattern as in the conventional example will be described.

The vowel patterns and consonant patterns thus separated are stored in vowel input patterns 4 and consonant input patterns 5, respectively.

When the recognition device is in the registration mode, the input patterns in the vowel input pattern storage section 4 and the consonant input pattern storage section 5 are stored in the vowel standard pattern storage section 6 and the consonant standard pattern storage section 7, respectively.

Next, in the recognition mode, the vowel similarity calculation section 8 outputs the number corresponding to the vowel standard pattern closest to the vowel input pattern to the output section 10;
Accordingly, the number corresponding to the consonant standard pattern closest to the consonant input pattern is output to the output unit 1°. The output unit 1o integrates the consonant recognition results and the vowel recognition results and outputs a monosyllable recognition result.

As explained above, according to the present invention, the cumulative function calculation unit 1
1, it is possible to easily limit the existence range of the vowel-consonant boundary, reduce the vowel-consonant boundary detection error, and improve the recognition rate.

Effects of the Invention As described above, according to the present invention, by calculating the cumulative function prior to detecting the vowel-consonant boundary, it is possible to easily limit the existence range of the vowel-7 consonant boundary, and to detect the vowel-consonant boundary. Detection errors can be kept within a small range.

This makes it possible to stabilize and improve the recognition rate, which has great practical effects.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a monosyllabic speech recognition device according to an embodiment of the present invention, FIG. 2 is an explanatory diagram of the operation of the same embodiment, FIG. 3 is a block diagram of a conventional monosyllabic speech recognition device, and FIG. 4 is an explanatory diagram of the operation of the conventional example, and Fig. 6 is an incorrect vowel of the conventional example.
It is an explanatory diagram showing an example of consonant boundary detection. 1: Input section, 2... Feature extraction section, 3
... Vowel-consonant boundary detection section, 4 ... Vowel input pattern storage section, 6 ... Consonant input pattern storage section, 6 ... Vowel standard pattern storage Part, 7...
. . . Consonant standard pattern storage section, 8 . . . Vowel similarity calculation section, 9 . . . Consonant similarity calculation section, 10.
. . . Output section, 11 . . . Cumulative function calculation section according to the present invention.

Claims (1)

[Claims] an input unit that converts input monosyllabic speech into an electrical signal;
A feature extraction unit extracts audio features from the output signal of the input unit, and a function is calculated by accumulating the changes in the feature amount of the audio extracted by the feature extraction unit over a certain period of time. a cumulative function calculation unit that sets the vowel-consonant boundary existence range of the monosyllabic speech to be around a point in time where the value obtained by dividing the value at the end of audio capture by a certain constant and the value at a certain point in the cumulative function are equal; , a vowel-consonant boundary detection unit that detects a vowel-consonant boundary within the vowel-consonant boundary existence range determined by the cumulative function calculation unit, and a vowel that stores the vowel pattern separated by the vowel-consonant boundary detection unit. an input pattern storage section, a consonant input pattern storage section that stores the consonant patterns separated by the vowel-consonant boundary detection section, a vowel standard pattern storage section that stores the vowel standard patterns, and a consonant standard pattern storage section that stores the consonant standard patterns. a consonant standard pattern storage section for storing a consonant standard pattern; a vowel similarity calculating section for calculating vowel similarity between the vowel input pattern storage section and the vowel standard pattern storage section; and a consonant standard pattern storage section and the consonant standard pattern storage section. a consonant similarity calculation unit that calculates a consonant similarity with a consonant similarity calculation unit; and an output unit that outputs a recognition result based on the output results of the vowel similarity calculation unit and the consonant similarity calculation unit. Syllable speech recognizer.