Speech Recognition: Acoustic, Phonetic and Lexical Knowledge
Annual progress rept. 1 Jul 1984-30 Jun 1985
MASSACHUSETTS INST OF TECH CAMBRIDGE RESEARCH LAB OF ELECTRONICS
Pagination or Media Count:
During this reporting period we continued to make progress on the acquisition of acoustic-phonetic and lexical knowledge. 1We completed development of a continuous digit recognition system. The system was constructed to investigate the use of acoustic-phonetic knowledge in a speech recognition system. The significant achievements of this study include the development of a soft-failure procedure for lexical access and the discovery of a set of acoustic-phonetic features for verification. 2 We completed a study of the constraints that lexical stress imposes on word recognition. We found that lexical stress information alone can, on the average, reduce the number of word candidates from a large dictionary by more than 80 percent. In conjunction with this study, we successfully developed a system that automatically determines the stress pattern of a word from the acoustic signal. 3 We performed an acoustic study on the characteristics of nasal consonants and nasalized vowels. We have also developed recognition algorithms for nasal murmurs and nasalized vowels in continuous speech. 4 We finished the preliminary development of a system that aligns a speech waveform with the corresponding phonetic transcription.
- Voice Communications