M. Pucher (Austria), Y. Huang, and Ö. Çetin (USA)
Speech Recognition, Latent Semantic Indexing
Latent Semantic Analysis (LSA) defines a semantic simi larity space using a training corpus. This semantic similar ity can be used for dealing with long distance dependen cies, which are an inherent problem for traditional word based n-gram models. Since LSA models adapt dynam ically to topics, and meetings have clear topics, we con jecture that these models can improve speech recognition accuracy on meetings. This paper presents perplexity and word error rate results for LSA models for meetings. We present results for models trained on a variety of corpora including meeting data and background domain data, and for combinations of multiple LSA models together with a word-based n-gram model. We show that the meeting and background LSA models can improve over the baseline n gram models in terms of perplexity and that some back ground LSA models can significantly improve over the n gram models in terms of word error rate. For the combina tion of multiple LSA models we did however not see such an improvement.
Important Links:
Go Back
