Title :
Speaker stress-resistant continuous speech recognition
Author :
Paul, Douglas B. ; Martin, Edward A.
Author_Institution :
Lincoln Lab., MIT, Lexington, MA, USA
Abstract :
Most speech recognizers are sensitive to the speech style and the speaker´s environment. This system extends an earlier robust continuous observation HMM IWR system to continuous speech using the DARPA-robust (multi-condition with a pilot´s facemask) database. Performance on a 207 word, perplexity 14 task is 0.9% word error rate under office conditions and 2.5% (best speaker) and 5% (4 speaker average) for the normal test condition of the database
Keywords :
speech recognition; DARPA robust database; continuous speech recognition; speaker stress resistant; Aircraft; Cognition; Databases; Degradation; Error analysis; Robustness; Speech recognition; Testing; Training data; Vocabulary;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1988. ICASSP-88., 1988 International Conference on
Conference_Location :
New York, NY
DOI :
10.1109/ICASSP.1988.196571
