Detailed Phonetic Labeling of Multi-language Database for Spoken Language Processing Applications
Final technical rept. Jan 2012-Jan 2015
STATE UNIV OF NEW YORK AT BINGHAMTON DEPT OF ELECTRICAL AND COMPUTER ENGINEERING
Pagination or Media Count:
The main objective of this research was to explore and refine methods for detailed phonetic labeling English or Russian and character level labeling Mandarin. Much of the work involved new front end signal processing methods designed to improve acoustic phonetic representations for speech. This resulted in recognition rates for TIMIT English of 74, among the highest reported in the literature. A complete character recognition system for Mandarin was developed and tested. Character recognition rates as high as 88 were obtained, using an approximately 40 training databases. For the case of Russian, a system for automatically converting Russian to morphemes, a kind of base syllable, was created and tested. A suite of tools for front end processing, automatic forced alignment, and a complete automatic recognition system are described.