Abstract:
Various hearing experiments and statistical analyses have been conducted concerning the relationships between individual and phonetic features in speech waves, and the following results have been obtained. (1) Statistical analysis of individual and phonetic information in speech waves shows that the latter is more significant than the former and that they have a significant effect on each other. (2) Hearing experiments using Japanese syllables extracted from continuous speech show that syllables used to perceive individuality are different from speaker to speaker. (3) Syllable identification tests using many speakers' voices have shown that people can adapt to a new speaker's voice after hearing just a few syllables, irrespective of the phonetic contents of the syllables. This means that people can generalize individual features extracted from a limited set of phonemes and apply this knowledge to other phonemes. The results from (1)--(3) imply that the same individual features are shared by several phonemes, although not all phonemes. (4) Other hearing experiments show that crucial phonetic information mainly exists in a short speech segment with a large spectral transition. Consequently, the simultaneous and automatic extraction of phonetic and individual information in transitional (dynamic) features of speech is an important research issue for speech information processing, including efficient speaker adaptation techniques in automatic speech recognition.