Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

Distinctive Feature Based SVM Discriminant Features for Improvements to Phone Recognition on Telephone Band Speech

Sarah Borys, Mark Hasegawa-Johnson

University of Illinois at Urbana-Champaign, USA

Support vector machines (SVM's) can be trained to classify manner transitions between phones and to identify the place of articulation of any given phone with high accuracy. The discriminant outputs of these SVM's can be used as input features for a standard ASR system. There is a significant improvement in correctness and accuracy using these SVM discriminant features when compared to an MFCC based recognizer of equal parameters.

Full Paper

Bibliographic reference.  Borys, Sarah / Hasegawa-Johnson, Mark (2005): "Distinctive feature based SVM discriminant features for improvements to phone recognition on telephone band speech", In INTERSPEECH-2005, 697-700.