Speaker Adaptation is a technique which is used to improve the recognition accuracy of Automatic Speech Recognition (ASR) systems. Here, we report a study of the impact of online speaker adaptation on the performance of a speaker independent, continuous speech recognition system for Hindi language. The speaker adaptation is performed using the Maximum Likelihood Linear Regression (MLLR) transformation of acoustic models. The ASR system was trained using narrowband speech. The efficacy of the speaker adaptation is studied by using an unrelated speech database as test data. The MLLR transform based speaker adaptation technique is found to improve the accuracy of the Hindi ASR system by 3%. It was also observed that the improvement in accuracy is dependent upon the recognition accuracy of the un-adapted system.

1. Rabiner, L.R. 1989. A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition. In Proceedings of the IEEE 77, 257—286.
2. Legetter, C.J. 1995. Improved acoustic modeling for HMMs using linear transformations. Ph.D. Thesis. University of Cambridge, (1995)
3. Leggetter, C.J., Woodland, P.C. 1995. Maximum likelihood linear regression for speaker adaptation of HMMs. Computer Speech & Language 9, 171–185.
4. Doh, S.-J. 2000. Enhancements to Transformation-Based Speaker Adaptation: Principal Component and Inter-Class Maximum Likelihood Linear Regression, PhD. Thesis, Carnegie Mellon University.
5. CMU sphinx – Speech Recognition Toolkit, http://www.cmusphinx.sourceforge.net.
6. Chan, A., et al. 2003. The Hieroglyphs: Building Speech Applications Using CMU Sphinx and Related
7. Samudravijaya, K. 2000. Hindi Speech Recognition. J. Acoustic Society of India 29(1), 385–393.
8. Sivaraman, G. et.al. 2011. Higher Accuracy of Hindi Accuracy of Hindi Speech Recognition Due to Online Speaker Adaptation, In Proceedings of ICTSM 2011, CCIS 145, 233 – 238.

Automatic Speech Recognition online speaker adaptation Maximum Likelihood Linear Regression (MLLR) Hindi Speech recognition