JPS62217298A - 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 present invention relates to a speech recognition device, and more particularly, to a recognition process and a dictionary creation process of the speech recognition device.

100% of voice interval detection in the voice recognition device
It is difficult to obtain perfect input, especially in noisy environments, as many devices have a variable threshold method.
Word-final consonant blocks are often not detected. Furthermore, in speaker-independent recognition methods, there are variations between individuals, and the power strength and duration of consonant blocks may be extremely different. In such a case, a certain specific block is missing, and due to the difference when creating a dictionary or during recognition, a correct dictionary cannot be created or erroneous recognition occurs.

Therefore, conventionally, when a certain speech block is missing or not depending on the speaker, a dictionary is prepared in advance for both the missing words and the non-missing words as words that are difficult to recognize. Multi-plate processing was performed. or.

When creating a dictionary, the numbers were simply added, which often resulted in poor accuracy of the dictionary.

Purpose The present invention was made in view of the above-mentioned circumstances.
In particular, this was done with the aim of improving the precision of the dictionary in the speech recognition device, thereby improving the speech recognition accuracy.

In order to achieve the above-mentioned object, the present invention provides means for extracting the characteristic amount of voice input from a microphone, means for detecting voice sections, and means for detecting voiced sections and silent sections within one word. In a speech recognition device that is equipped with means for segmenting, means for registering an additive dictionary in block units based on the silent sections, and means for performing recognition processing based on the silent sections, consonants that are relatively likely to be omitted when creating a dictionary are used. A word having a block, etc. is specified in advance, and during recognition, the degree of similarity is calculated for both cases in which the block is missing and not, and the higher one is taken as the degree of similarity for that word. It is characterized by Hereinafter, the present invention will be explained based on examples.

FIG. 1 is an electrical block diagram for explaining an embodiment of the present invention. In the figure, 1 is a microphone, 2 is a preprocessing section, 3 is a feature extraction section, and 4 is a voice section detection section.

5 is a recognition processing and dictionary creation unit; 6 is an identifier check unit;
Reference numeral 7 denotes a word dictionary, and as an example, a case will be described in which it is applied to a speaker-independent recognition device using the BTSP method and focusing on silent sections.

FIG. 2 is a diagram for explaining an example of a recognition method that focuses on a silent section. This recognition method that focuses on a silent section uses a silent section B that occurs in a vocal section A as a node, and blocks the sound on both sides of the silent section B. This is to match C comrades and perform pattern matching. However, for example, in the illustrated word ``stop,'' depending on the speaker, 1 p' may be omitted, as shown in FIG. For example, if ``pu'' is missing during input, the input 1st stroke'' will be recognized as the dictionary ``stop'', resulting in erroneous recognition.

The present invention was made in order to prevent the above-mentioned misrecognition.In the present invention, as mentioned above, an identifier is attached to a word in advance at the time of dictionary creation, and for words with this identifier, Recognition calculations are performed twice, one for when a word is attached to a block at the end of a word in the dictionary, and another for when it is not attached, and the one with the greater degree of similarity is taken as the degree of similarity for that word. This type of missing endings is especially common in English and German, and the same can be said for things like '5kip' and 'Up'.Also, if one ending is missing, or if it is at the beginning of a word, you can change the identifier accordingly. The method of setting these two times during recognition can be easily determined.

Next, a case will be described in which the present invention is applied when creating a dictionary. As in the case of recognition, the dictionary is created by focusing on silent sections, so if a voiced block in a word is missing, the quality of the created dictionary deteriorates. Further, when creating a dictionary, for example, when creating an additive dictionary three times, there is a method of calculating and checking the degree of similarity. Therefore, when calculating the similarity, two similarity calculations are performed, assuming that the block is missing and when it is not. By excluding that block and adding only other blocks to the dictionary, it is possible to create an accurate dictionary.

FIG. 4 is a diagram showing an example of such a case, where ■ is the first verification and ■ is the second verification. In addition, adding an identifier in advance as mentioned above can be easily provided if, for example, a dictionary is created and supplied, but if the device is used to create the identifier, the identifier cannot be created automatically. must not. In order to do this, if the result of performing the above two calculations to calculate the similarity is that the similarity is higher when the block is missing, then the input may be missing, so we need to add that identifier to the word. It would be a good idea to add it. In this way, when creating a dictionary, the similarity calculation may be performed several times during recognition according to the identifier, and the one with the highest similarity may be used as the recognition result.

Effects As is clear from the above explanation, according to the present invention, it is possible to accurately create a dictionary and recognize words even for words whose blocks are frequently missing.

[Brief explanation of drawings]

FIG. 1 is an electrical block diagram for explaining an embodiment of a speech recognition device according to the present invention, and FIG. 2 is a diagram for explaining an example of a recognition method focusing on silent intervals. FIG. 3 is a diagram for explaining the drawbacks of the conventional recognition method, and FIG. 4 is a diagram for explaining one method for creating a dictionary. DESCRIPTION OF SYMBOLS 1... Microphone, 2... Preprocessing section, 3... Feature extraction section, 4... Speech section detection section, 5... Recognition processing and dictionary creation section. 6...Identifier check unit, 7...Word dictionary. Figure 1 Figure 2 Figure 3! ! I evening 4 @

Claims (2)

[Claims] (1) A means for extracting the characteristic amount of the voice input from the microphone, a means for detecting the voice interval, a means for separating the voiced interval and the silent interval within one word, and In a speech recognition device that is equipped with a means for registering an additive dictionary and a means for performing recognition processing based on silent intervals, when creating a dictionary, words having consonant blocks etc. that are relatively easy to be omitted are specified in advance, and then recognized. A speech recognition device comprising a pattern matching method that sometimes calculates the degree of similarity for an input pattern both when that block is missing and when it does not, and sets the higher one as the degree of similarity for the word. (2) The speech recognition device according to claim (1), characterized in that when creating the dictionary, dictionary addition is performed while determining similarity.