SVM Based Speaker Selection Using GMM Supervector for Rapid Speaker Adaptation
In this paper, we propose a novel method for rapid speaker adaptation called speaker support vector selection (SSVS). By taking gaussian mixture model (GMM) as speaker model, the speakers acoustically close to the test speaker are selected .Different from other selection method, just computing the likelihood between models, we utilizing support vector machines (SVM) to obtain a ‘more optimal speaker subset’. Such selection is dynamically determined according to the distribution of reference speakers close the test. Furthermore, a single-pass re-estimation procedure conditioned on the selected speakers is shown. This adaptation strategy was evaluated in a large vocabulary speech recognition task. The presented method improves the relative accuracy rates by 13% compared to the baseline system.
KeywordsSupport Vector Machine Hide Markov Model Gaussian Mixture Model Speaker Identification Hide Markov Model Model
Unable to display preview. Download preview PDF.
- 3.Sankar, A., Beaufays, F., Digalakis, V.: Training data clustering for improved speech recognition. In: Proc. Eurospeech, pp. 502–505 (1995)Google Scholar
- 4.Huang, C., Chen, T., Chang, E.: Speaker Selection Training for Large Vocabulary Continuous Speech Recognition. In: Proc. ICASSP (2002)Google Scholar
- 9.Gunn, S.R.: Support vector machines for classification and regression. Technical Report Image Speech and Intelligent Systems Research Group, University of Southampton (1997)Google Scholar
- 10.Schmidt, M., Gish, H.: Speaker identification via support vector classifiers. In: Proc. ICASSP, pp. 105–108 (1996)Google Scholar