| 摘要 |
A very large vocabulary isolated word speech recognition system is provided, wherein speech signals are received into a processor. A phonetic dictionary is formed from a baseform dictionary by applying up to four sets of phonological rules to generate phonetic spelling variations for each word. The spelling variations may account for acoustic variations comprising dialect, phonological processes of language, acoustic-phonetic processes of language, and overpronunciation. The received speech signals are processed using a speech recognition system comprising the phonetic dictionary and at least one phoneme set. In one embodiment, the phoneme set comprises a single phoneme to account for stop closures and glottal stops. Moreover, the phoneme set comprises a reduced mid-central unstressed vowel and a reduced high-central unstressed vowel. Furthermore, the speech recognition system is produced by generating a number of phoneme models, some of which are shared among a number of phonemes. Output signals are generated that are representative of the received speech signals. |
