Sixth International Conference on Spoken Language Processing
Phoneme recognition is a difficult task in speech recognition as it is variable in length and its acoustic properties change due to co-articulation and variation in dialects. The performance of the speech recognition system is heavily based on features extracted for the phonemes. The conventional technique of Short Time Fourier Transform (STFT) has a serious limitation in resolving the stop (plosive) sounds. This shortcoming can be overcome by using the multi-resolution capability of Wavelet Analysis. In this paper we perform a comparative study of Discrete Wavelet Transform (DWT) and Wavelet Packet (WP) for new dynamic features extraction of phonemes.
Bibliographic reference. Farooq, Omar / Datta, Sekharjit (2000): "Dynamic feature extraction by wavelet analysis", In ICSLP-2000, vol.4, 696-699.