JPH0574838B2 - - Google Patents

Description

[Detailed description of the invention]

[Industrial Application Field] The present invention relates to a method for automatically calculating distance between words,
This is effective when used to pre-evaluate the suitability of a target word set for recognition by a speech recognition device. [Prior Art] Some speech recognition devices have a predetermined group of words (referred to as a recognition target word set) that may be input by voice.In order to improve the recognition rate of input words of such speech recognition devices, It is important to ensure that there are no pronunciations (pronunciations) similar to the word set. The words themselves included in the word set are determined by the purpose of use of the device and are expected to be difficult to change, but there is no particular problem in changing the pronunciation of the words, so the pronunciation of each word in the word set is not confusing. It is best to select as follows. Sometimes it's easy to tell whether something is confusing or not.
For example, the number ``7'' can be read as ``shichi'' or ``nana,'' but when pronounced as ``shichi,'' it is confusingly pronounced as the number ``1,'' or ``ichi,'' and both are easily misrecognized. Therefore, if "7" is pronounced as "nana", it will be clearly distinguished from "ichi", and this is known from experience. However, as the number of words targeted for speech recognition increases, it becomes difficult to tell which words are confusing, and when one finds a confusing word and substitutes another pronunciation, it becomes confusing with the other word. There are also problems. Therefore, for the word set to be used for speech recognition, for words that have two or three different pronunciations, select an appropriate one and fix the pronunciation of the word set, and then try actually performing speech recognition to see if there are any misrecognitions. We use a cut-and-try method to find an appropriate set of words. However, this method requires an extremely large amount of time and labor. Therefore, rather than conducting an actual speech recognition test, it is effective to check the pronunciation of each word in the word set in advance, while it is still in the character stage. Considering that the problem of confusion occurs between any two words in a word set, it is necessary to investigate the degree of confusion between every word in the word set and all the remaining words to eliminate the risk of misrecognition. It is effective to reject a word set if there is a certain word pair (pronunciation). Word distance is a numerical expression of the degree of similarity and dissimilarity between words. DP matching method (Velichko et al.Int.J.
Man-Machine Studies, vol.2, p223, 1970)
Now, find this distance between words as the distance between character strings. Simply put, the local distance between each syllable of two words is accumulated, and there are various possible combinations of these syllables, and the cumulative value differs depending on the combination, but the minimum value among them is taken as the distance between words. be. [Problem to be solved by the invention] However, this DP method does not reflect the number of syllables in each word. For example, word A with 2 syllables
Suppose that the distance between and word B with 3 syllables is equal to the distance between word C with 2 syllables and word D with 2 syllables, then the distance between words C and D in the pair of words A and B is It is said that the false recognition rate for the pairs is the same, but
From experience, the recognition rate for sets A and B, which have different numbers of syllables, should be higher than the recognition rate for sets C and D, which have the same number of characters. This point is not reflected in distance calculations using the conventional DP matching method. The present invention compensates for the insufficiencies of the DP method described above,
This is an attempt to make the prior evaluation of recognition target words more practical. [Means for Solving the Problems] The present invention provides a method for calculating the distance between each word in a word set to be speech recognized, which includes a step of decomposing each word into individual syllables, and a step of dividing each word into individual syllables. The step of calculating the distance to a word with the number of syllables N using the DP matching method, and the distance obtained, if the two words have the same number of syllables, the distance is smaller than if the two words have the same word length. In the case where the distance to the end point determined by the DP method is divided by the word length, the normalized distance is
This method is characterized by a step of determining the distance between words by multiplying it by a normalization constant that makes corrections that satisfy three conditions: the distance is always symmetrical with respect to the two distances. Next, the structure and operation will be explained in detail with reference to embodiments. [Embodiment] Fig. 1 is a schematic diagram of a system in which two words A and B written in kana are decomposed into syllables in processing blocks 1 and 2, and the distance between the words is determined in processing block 3 by the DP method. 4 is a correspondence table between kana and syllables used when decomposing into syllables, and 5 is a phoneme distance matrix used when calculating distances. The kana notation of words A and B is 50 sounds (46 sounds), voiced sounds,
It consists of nasal duplication, semi-voiced consonance, consonant consonant, consonant consonant, consonant consonant, consonant consonant, loanword "sui", "tei", etc., and any combination of one or more of these long consonants. Further, the syllables divided by processing blocks 1 and 2 consist of consonants and vowels, where the vowels consist of a, i, u, . . . , etc., and the consonants consist of s, k, t, . . . . A part of the romaji notation table for kana is shown below.

【table】 a

Claims (1)

[Scope of Claims] 1. A method for calculating the distance between each word of a word set to be speech recognized, comprising the steps of decomposing each word into individual syllables, and dividing a word with M syllables into a word with N syllables. distance
If two words have the same number of syllables, the distance is smaller than if the two words have the same number of syllables.
If two words have the same word length, DP
The normalized distance is the distance to the end point calculated by the word length divided by the word length, and the distance is always symmetric about the two distances. 1. A method for automatically calculating distances between words, comprising the step of calculating distances between words.