Human emotions can be recognized through speech analysis. One main problem of this discipline is the lack of databases with a sufficient number of patterns for a correct learning. This fact makes generalization in the learning process be more difficult. One possible solution is the creation of new virtual patterns, enlarging the training set. In order to carry out this enlargement, we modify the average pitch by using the technique known as Pitch Synchronous Overlap and Add combined with resampling, that allows to change the average pitch without altering neither the pitch variations nor the speech rate. Therefore, the emotion in the utterance is unaltered. Results over the original test set show that it is possible to achieve a significant reduction in the generalization effects with the proposed creation of new virtual training patterns.
Mohino-Herranz, Inma; Sánchez-Hevia, Héctor A.; Gil-Pita, Roberto; Rosa-Zurera, Manuel
Affiliation: Universidad de Alcalá, Alcalá de Henares, Madrid, Spain
AES Convention: 136 (April 2014) Paper Number: 9037
Publication Date: April 25, 2014
Subject: Signal Processing
No AES members have commented on this paper yet.
If you are not yet an AES member and have something important to say about this paper then we urge you to join the AES today and make your voice heard. You can join online today by clicking here.