First International Conference on Spoken Language Processing (ICSLP 90)
This paper describes a new technique for isolated word recognition that uses both pitch information and spectral information. Words with similar phonetic features tend to be misrecognized in conventional methods which use only spectral information, even if their phonemes are accented differently. Many phonetically-similar Japanese words are classified by pitch patterns. This paper introduces a measure of the pitch pattern distance. A pitch pattern template is produced by averaging pitch patterns obtained from a set of words which have the same accent pattern. A measure for word recognition is proposed, based on a combination of the pitch pattern distance and the phonetic likelihood. Speaker-dependent word recognition experiments were carried out using 216 Japanese words uttered by five male and five female speakers. The proposed measure reduces the recognition error rate by 40% compared with the conventional phonetic likelihood.
Bibliographic reference. Takahashi, Satoshi / Matsunaga, Shoichi / Sagayama, Shigeki (1990): "Isolated word recognition using pitch pattern information", In ICSLP-1990, 553-556.