Automatic labelling of foreign-accented speech
Rene O. Arechiga · 2008
This paper describes the automatic labelling of a Spanish-accented corpus of Mexican students of English as a second language. Some preprocessing is done prior to perform forced alignment with a hidden Markov models-based speech recognizer trained with the TIMIT database. Maximum likelihood linear regression adaptation is used to improve the phone segmentation. The labelling results are fair, but still not good enough for practical applications.