| 摘要 |
<p>The method includes the preliminary steps of: determining and storing, for each word in the vocabulary, at least one corresponding "phonetic model" constituted by a sequence of phonetic symbols corresponding to the phonemes of the word; and determining and storing, for a limited set of basic phonemes or "prototypes" spoken by the speaker, the associated values assumed by predetermined characteristic parameters indicative of the energy and spectral composition of that phoneme. Each time it is carried out, the method then includes the steps of: generating an electrical sound signal corresponding to each isolated word spoken by the speaker; sampling the electrical sound signal and determining the values assumed by the characteristic parameters for each sample, for each sample, comparing the corresponding values of the characteristic parameters with the previously-stored values of the same parameters of the basic phonemes or "prototypes"; determining, on the basis of this comparison, a sequence or "string" of phonetic symbols constituting an approximate representation of the phonemes contained in the isolated word spoken, and comparing the phonetic "string" with all the phonetic "models" of the vocabulary and selecting, by means of a dynamic programming algorithm, a small number of models ("candidate" models) whose characteristics best correspond to those of the "string".</p> |
