Click here to register.

Google Summer of Code Ideas Page

Re: how to get more voice samples?
User: kmaclean
Date: 1/22/2010 8:11 pm
Views: 74
Rating: 8

From Acoustic Model Clustering Based on Syllable Structure by Izhak Shafran & Mari Ostendorf:

Recognizing conversational speech has proved to be more challenging than read speech for automatic speech recognition (ASR) systems. For the best systems reporting results on the 1999 DARPA Broadcast News benchmark tests, word error rates on the spontaneous speech portion of the test set (14-16%) were nearly double those on the baseline condition of planned recordings (8-9%) (Pallett et al., 1999). [...]

The degradation in performance may be due to many factors such as channel e ffects, variability in speaking rate and dialect of speakers, less careful pronunciation, loosely structured language, and the presence of disfluencies. In 1996, (Weintraub et al., 1996) demonstrated that a large part of the degradation is related to acoustic variation associated with speaking style.

Re: how to get more voice samples?
User: Mariane
Date: 1/23/2010 8:34 am
Views: 2349
Rating: 7

Thank you for the reference.