Mañana, JUEVES, 24 DE ABRIL, el sistema se apagará debido a tareas habituales de mantenimiento a partir de las 9 de la mañana. Lamentamos las molestias.
Feature learning with raw-waveform CLDNNs for Voice Activity Detection
Entity
UAM. Departamento de Tecnología Electrónica y de las ComunicacionesPublisher
International Speech and Communication AssociationDate
2016-09-12Citation
10.21437/Interspeech.2016-268
Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH. International Speech and Communication Association, 2016. 3668-3672
ISSN
2308-457XDOI
10.21437/Interspeech.2016-268Editor's Version
http://dx.doi.org/10.21437/Interspeech.2016-268Subjects
Feature extraction; Network architecture; Speech communication; Speech processing; Deep neural networks; Frequency modeling; Long short term memory; Model architecture; Pre-processing step; Proposed architectures; Speech recognition systems; Voice activity detection; Speech recognition; TelecomunicacionesRights
© 2016 ISCAAbstract
Voice Activity Detection (VAD) is an important preprocessing step in any state-of-the-art speech recognition system. Choosing the right set of features and model architecture can be challenging and is an active area of research. In this paper we propose a novel approach to VAD to tackle both feature and model selection jointly. The proposed method is based on a CLDNN (Convolutional, Long Short-Term Memory, Deep Neural Networks) architecture fed directly with the raw waveform. We show that using the raw waveform allows the neural network to learn features directly for the task at hand, which is more powerful than using log-mel features, specially for noisy environments. In addition, using a CLDNN, which takes advantage of both frequency modeling with the CNN and temporal modeling with LSTM, is a much better model for VAD compared to the DNN. The proposed system achieves over 78% relative improvement in False Alarms (FA) at the operating point of 2% False Rejects (FR) on both clean and noisy conditions compared to a DNN of comparable size trained with log-mel features. In addition, we study the impact of the model size and the learned features to provide a better understanding of the proposed architecture.
Files in this item
Google Scholar:Zazo Candil, Rubén
-
Sainath, Tara N.
-
Simko, Gabor
-
Parada, Carolina
This item appears in the following Collection(s)
Related items
Showing items related by title, author, creator and subject.