JPH0766274B2 - Word speech recognizer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a speech recognition technique, and more particularly to improvement of a word speech recognition technique.

(Prior art and its problems) With a word voice recognition device, a recognition word is set according to the application of the user, each word of the word is registered as a standard pattern, and recognition is performed using this. However, the conventional word voice recognition device is prone to erroneous recognition depending on the set word. This is because there may be words that are similar to each other among the words to be recognized.

The present invention is a means for allowing a user to set a word name that a user wants to use as a plurality of word groups that can be substituted, and for automatically determining a combination that is most unlikely to cause misrecognition in these word groups. The object of the present invention is to realize a voice recognition device in which recognition errors are unlikely to occur by incorporating the above into the device.

(Means for Solving Problems) Further, according to the invention of the present application, when setting a vocabulary to be recognized, a word group consisting of a plurality of words that can be mutually replaced for each of the target word functions (alternative (Group of words) and determining a vocabulary by selecting one word from each of the alternative word groups. Is performed by using the distance between words calculated by the phonetic description of the word.

(Operation) The basic principle of the present invention will be described below. Let K be the type of word required by the user. Here, this will be called that there are K types of categories. The user gives a possible word name as a substitute for each word. That is, the user gives a group of possible words for each category. On the other hand, the distance D (X, Y) between words is considered as a measure of the likelihood of misrecognition between words. Here, X and Y represent words.

The problem is to select the words from the group of words belonging to each category so that the distance between the selected words becomes as large as possible. In terms of misrecognition, it is necessary to avoid the presence of word pairs with small interword distances. For this purpose, the words may be selected so that the minimum value of the distances between the selected words becomes the maximum. When the number of selected combinations is small, the inter-word distance may be calculated for all the combinations, but the processing amount increases remarkably as the number of words increases. Here, the following method, which is not optimal but more effective than keeping the distance between words as large as possible, will be described.

The basic procedure is to sequentially delete words having a small interword distance from other words from the alternative word group of each category, and finally leave one word for each category. Specifically, it is as follows.

(1) For each word W in the alternative word group of all categories, calculate the distance from all the words other than the category to which it belongs (excluding the words that have already been deleted), and calculate the minimum value of these distances. Let d (w).

(2) W giving the minimum value of the obtained d (w)
Select (= w ^* ) and delete it. However, when w ^* is the only word belonging to that category, w is selected instead of w ^* , which gives the second smallest value.

The following method is shown here as a method of determining D (X, Y).

Phonetic description of a word, that is, a phoneme (or syllable) of a word
This is a method of using the columns of. Phonemes (or syllables)
If the distance between words is defined in advance based on the knowledge of phonetics, the distance between words can be calculated by using this distance between phonemes instead of the distance between patterns in the normal DP matching method. It That is, the phonetic descriptions of the two words to be compared are A = {a ₁ ... A _M }, B =
If {b ₁ ... b _N } and the phoneme X, Y distance is d (x, y), the word distance D (A, B) can be calculated by the following recurrence formula.

When initial condition g (m, n) = 0, m = 0, n = 0 = ∽ (m> 0, n
= 0) or (m = 0, u> 0) m = 1, ..., M; n = 1, ..., N D (A, B) = g (M, N) / (M + N) (Embodiment) FIG. 1 shows an embodiment of an apparatus realizing the present invention. It is a block diagram showing. Reference numeral 1 is a pattern and stores the phonetic description of a word group belonging to each category. Reference numeral 2 is an inter-pattern distance calculation unit, which is a word group W ^K (1) ... W ^K (J belonging to each category k in the pattern buffer 1
A group {d (j ₁ , j ₂ )} of distances between (K)) is calculated and stored in the inter-pattern distance buffer 3. Here, d (j ₁ , j ₂ ) is the inter-pattern distance between the words W ^K (j ₁ ) and W ^K (j ₂ ) and is calculated by the method described above. Reference numeral 4 is a pattern selection unit, which reads the contents of the distance buffer 3 described above, executes the above-described procedure for sequentially deleting words, and finally selects a word to be used as a standard pattern and outputs the selection result. The recognition unit 5 performs standard pattern registration for the word selected above,
Word recognition is performed using this. Any word recognition can be used as long as it is based on matching with a standard pattern.

(Effects of the Invention) As described above, according to the present invention, it is possible to incorporate a means for automatically determining a combination that is most unlikely to cause misrecognition in a plurality of word groups into the device, and to improve recognition accuracy. It is possible to realize a high-quality voice recognition device.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment according to the present invention, in which 1 ... Pattern buffer, 2 ... distance calculator,
3 ... Distance buffer, 4 ... selection unit, 5 ... recognition unit.

Claims (1)

[Claims] 1. When setting a vocabulary to be recognized,
Includes a function that defines a word group consisting of multiple words that can be substituted for each target word function as an alternative word group, and selects a word from each alternative word group The word speech recognition apparatus characterized in that the word selection in the vocabulary determination means is performed by using the inter-word distances calculated by the phonetic description of the words in the alternative word group.