JPS61204699A - Reference pattern registration 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 relates to a standard pattern registration method used for speech recognition.

In general, voice recognition involves registering a standard voice pattern in advance.
The similarity or distance between unknown input cover patterns and their similarity is often determined, and the standard pattern with the maximum similarity (minimum distance) is often used as the recognition result. As the number of registered standard patterns increases, the amount of memory increases. Therefore, there is a method in which only patterns that are basic units of speech are registered, and when recognition is performed, these are synthesized and used as a speech standard pattern. This basic unit is CV (consonant-vowel), CVC, VC
V, etc. are being considered, but the vowel parts in these basic units have different patterns registered even though they are the same vowel. For example, in CV units, 1ta1 and 1ka1 are created separately, and the patterns of the vowel part 1al are different. ltal
When trying to recognize the words lgal and lkallgal, it is sufficient to be able to distinguish only from the difference between ltl and lkl, but the difference in the vowel pattern following the consonant is also reflected in the similarity (distance) calculation. Moreover, the duration of 1tl and lkl is shorter than that of vowels, making it difficult to understand the characteristics;
The similarity (distance) of lal may have a greater effect on the overall similarity (distance) than that of lkl, which has the disadvantage of causing erroneous recognition.

The present invention has been made in view of the above-mentioned circumstances.
In particular, the basic units of audio (CV, VCV).

The purpose of this invention is to provide a standard pattern that can reduce the storage amount of standard patterns and further reduce erroneous recognition in speech recognition using concatenated standard patterns (CvC).

1. In order to achieve the above object, the present invention is based on the basic unit of speech (
This method is characterized by using a common pattern for the same vowel part in a basic unit in a standard pattern registration method by concatenating CV, VCV, CVC). Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a configuration diagram for explaining one embodiment of the present invention. In the figure, ■ is a C (consonant part) standard pattern part, 2 is a V (vowel part) standard pattern part, and 3 is a C2 ■ connection. 4 is a CV standard pattern part, 5 is a word label dictionary part, 6 is a CV connection part, and 7 is a word standard pattern part. A common vowel part (V) standard pattern is used for the same vowel part.

Hereinafter, with reference to FIG. 1, a method of registering a word standard pattern by concatenation of CV@ positions will be explained. here,
In CV syllables, we will consider C (consonant part) and ■ (vowel part) with the midpoint of the transition from consonant to vowel as the boundary. Good too.

In addition, the C standard pattern will be created for each subsequent vowel. These C1■ standard patterns are concatenated to create a C■ standard pattern, and the CV patterns are concatenated to form a word standard pattern by referring to the label of the word to be the standard pattern.

FIG. 2 shows the l taga 1. according to the present invention. lkag
This is a diagram for explaining an example of creating a word standard pattern for al. In the diagram, A is a C standard pattern, B is a ■ standard pattern, D is a Cv standard pattern, E is a word standard pattern, and the following vowel is lal. There is one tallkal, Igal c
v After creating the standard pattern, further 1ta1 and Iga
1, 1kal and Igal are concatenated to form a standard pattern. ltaga l and 1kag created in this way
The standard pattern for al is a pattern that differs only in the ltl and lkl parts.

Results As is clear from the above description, according to the present invention, differences between standard patterns can be clearly created, and patterns can be accurately discriminated in speech recognition.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram for explaining one embodiment of the present invention,
FIG. 2 is a diagram for explaining an example of creating a word standard pattern according to the present invention. 1...C (consonant part) standard pattern part, 2...■ (vowel part) standard pattern part, 3...C, V connection part, 4...
・C■Standard pattern section, 5...Word label dictionary section, 6
...CV connection part, 7...word standard pattern part.

Claims (3)

[Claims] (1) A standard pattern registration method by concatenating basic units of speech (CV, VCV, CVC), characterized in that a common pattern is used for the same vowel part in the basic units. (2) The basic units of audio (CV, VCV, CVC)
Claim (1) characterized in that the constant pattern section of the vowel (V) is regarded as the vowel part, and the crossing section preceding or following the constant vowel part is included in the consonant part. Standard pattern registration method described in section. (3) The standard pattern registration method according to claim (2), wherein the vowel part pattern is created by a series of one frame data having parameter values specific to the vowel.