Multilingual MLP features for low-resource LVCSR systems
Authors:
- 1 March 2012
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 4269-4272
Abstract
We introduce a new approach to training multilayer perceptrons (MLPs) for large vocabulary continuous speech recognition (LVCSR) in new languages which have only few hours of annotated in-domain training data (for example, 1 hour of data). In our approach, large amounts of annotated out-of-domain data from multiple languages are used to train multilingual MLP systems without dealing with the different phoneme sets for these languages. Features extracted from these MLP systems are used to train LVCSR systems in the low-resource language similar to the Tandem approach. In our experiments, the proposed features provide a relative improvement of about 30% in an low-resource LVCSR setting with only one hour of training data.Keywords
This publication has 8 references indexed in Scilit:
- Multilingual acoustic modeling for speech recognition based on subspace Gaussian Mixture Models2010
- A study on multilingual acoustic modeling for large vocabulary ASR2009
- Training and adapting MLP features for Arabic speech recognition2009
- Probabilistic and Bottle-Neck Features for LVCSR of Meetings2007
- Combination of Acoustic Classifiers Based on Dempster-Shafer Theory of Evidence2007
- Tandem connectionist feature extraction for conventional HMM systems2002
- SWITCHBOARD: telephone speech corpus for research and development1992
- Perceptual linear predictive (PLP) analysis of speechThe Journal of the Acoustical Society of America, 1990