JPS63149697A - 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 Industrial Application] The present invention relates to a speech recognition device that performs speech recognition by comparing a speech pattern with a standard pattern. The present invention relates to a speech recognition device that is expanded to.

[Conventional technology]

2. Description of the Related Art Conventionally, a speech recognition device is known that performs speech recognition by comparing the contents of a standard pattern memory in which predetermined standard patterns are registered with an input speech pattern.

[Problem that the invention seeks to solve]

The above-mentioned speech recognition device requires the standard pattern memory to operate at high speed, which makes the standard pattern memory expensive.As a result, in the speech recognition device that requires economy, only a small capacity standard pattern memory is required. Cannot be implemented. Therefore, there is a problem that there are few recognized words.

[Means for Solving the Problems] According to the present invention, a standard pattern includes a speech pattern memory in which a speech pattern to be recognized is temporarily stored, a resident part in which monosyllables are stored, and an overlay part. a memory, an expanded memory in which a plurality of words are grouped and stored based on the monosyllables, and a comparison section that compares the contents of the speech pattern memory and the contents of the standard pattern memory; The contents of the voice pattern memory are compared with the contents of the resident section in the section, and one group of the extended memory is transferred to the overlay section based on the comparison result, and the contents of the overlay section and the voice pattern are transferred to the overlay section. A speech recognition device is obtained in which speech recognition is performed by comparing the content of the memory with the content of the memory in the comparison section.

〔Example〕

Next, the present invention will be explained with reference to examples.

FIG. 1 is a block diagram showing an embodiment of a speech recognition device according to the present invention. The audio pattern input from the microphone 11 is amplified by the amplifier 12 and then digitized by the AD converter 13.

It is stored in the voice pattern memory 15 by the control unit 101 via the filter (FIL) 14.

The standard pattern memory 17 is composed of a resident section and an overlay section, which will be described later.
8 is written as appropriate under the control of the control unit 101. Resident part of standard pattern memory 17 and extended memory 1
8, the speaker's voice pattern is registered in advance as a standard pattern. Then, the comparison section 16 compares the contents of the voice pattern memory 15 and the contents of the standard pattern memory 17 word by word under the control of the control section 101. In FIG. 1, the solid lines represent data lines and the dotted lines represent signal lines for control.

FIGS. 2(a) and 2(b) show the contents of the standard pattern memory and the expanded memory, respectively. As shown in Figure 2 (a), the resident area of the standard pattern memory contains the monosyllable “a”.
Standard patterns from ``wa'' to ``wa'' are stored. on the other hand,
Figure 2 (b) K - As shown in the extended memory, word sound patterns are stored in groups in the order of ``A'' and ``I'', such as a group of word sound patterns with "a" and a group of word sound patterns with "i". has been done.

Suppose that speaker 0 has now uttered the word "Onsei."

This voice is from microphone 11. Multiplier 12° AD converter 1
3, the signal is divided into frequency components by a filter 14, and stored in a voice pattern memory 15. This memory 1
When the control unit 101 learns that the data has been stored in 5, the comparison unit 16
The contents of the resident part of the standard pattern memory 17 and the contents of the voice pattern memory 15 are compared using DP matching, aligning the beginnings of the contents of the resident part and the contents of the voice pattern memory. By the way, the content of the voice pattern memory is the word "Onsei", and the resident part is "A".
, "Ij, etc., so words naturally have longer pattern lengths, but by aligning the beginnings as described above, it is possible to compare the contents of the resident part with the contents of the speech pattern memory. .Patterns from the beginning of the pattern "Onsei" to the length of "A", "I", etc. are compared in the comparison section, and the degree of similarity is obtained as a number. ``Onsei'' stored in the voice pattern memory as shown in Figure 3.
The first "o" and the permanent part "a", "i"...
Compare with "wa" one by one. Then, the control unit 101
As a result of matching, when it is learned that "O" is the closest, it is determined that "O" is the closest one in the contents of the extended memory 18, as shown in FIG.
Standard pattern memory 1 contains the voice patterns of groups marked with
Transfer to the overlay part of No.7. After the transfer is completed, the comparing section again compares the contents of the overlay section of the standard pattern memory 17 with the contents of the audio pattern memory 15, selects the one that is close to the contents of the standard pattern memory from among the contents of the overlay section, and stores the code in the register. 19, and recognition is performed.

〔Effect of the invention〕

As explained above, the speech recognition device according to the present invention is provided with a standard pattern memory and an extended memory, and a part of the content of the extended memory is transferred to the standard pattern memory during speech recognition. By using a large-capacity, low-cost memory as the extended memory and a small-capacity, high-priced memory as the standard pattern memory, an overall low-cost speech recognition device with a large number of phrases can be obtained.

[Brief explanation of the drawing]

FIG. 3 is a diagram for explaining comparison by matching. 11...Microphone, 12...Amplifier, 13...A/
D converter, 14... Filter (F engineering L), 15...
Audio pattern memory, 16... Comparison section, 17... Standard pattern memory, 18... Expansion memory, 19...
Register, 101...control unit. Figure 2 Figure 3 Mouth

Claims (1)

[Claims] 1. A speech pattern memory in which speech patterns to be recognized are temporarily stored, a standard pattern memory comprising a resident section and an overlay section in which monosyllables are stored, and a plurality of words stored in groups based on the monosyllables. and a comparing section for comparing the contents of the voice pattern memory and the contents of the standard pattern memory, and the comparing section compares the contents of the voice pattern memory and the contents of the resident section. Then, one group of the extended memory is transferred to the overlay section based on the comparison result, and the comparison section compares the contents of the overlay section and the contents of the voice pattern memory to perform voice recognition. A voice recognition device characterized by: