摘要 |
PURPOSE: A speech recognition method and a device thereof are provided to combine the smallest pair including the learning slump data among distance values between triphones. CONSTITUTION: A phoneme distance value calculating unit(409-3) calculates mono-phone distance values between phonemes within phoneme series about a voice signal. If a learning slump data exists in triphone data about the phoneme series, a clustering unit(409-4) adds mono-phone distance values of context among the calculated mono-phone values. The clustering unit calculates distance between triphones. The clustering unit combines the smallest pair among distance values between the calculated triphones.
|