This report proposes and compares a number of tandem-like feature extraction schemes. The proposed schemes use relative phone posteriors as condence measures estimated from the MLP outputs directly or using Gamma function. The analysis of variances shows that the proposed tandem-like features discriminate better between phone classes than the conventional tandem features. But these capabilities are lost when the complexity of the model (number of gaussians) increases. Evaluations are conducted on TIMIT database.
Confidence measures for tandem connectionist feature extraction
Research report RR-06-166
© EURECOM. Personal use of this material is permitted. The definitive version of this paper was published in Research report RR-06-166 and is available at :
PERMALINK : https://www.eurecom.fr/publication/1967