Correlated Bigram LSA for Unsupervised Language Model Adaptation

We present a correlated bigram LSA approach for unsupervised LM adaptation for automatic speech recognition. The model is trained using efficient variational EM and smoothed using the proposed fractional Kneser-Ney smoothing which handles fractional counts. We address the scalability issue to large training corpora via bootstrapping of bigram LSA from unigram LSA. For LM adaptation, unigram and bigram LSA are integrated into the background N-gram LM via marginal adaptation and linear interpolation respectively. Experimental results on the Mandarin RT04 test set show that applying unigram and bigram LSA together yields 6%

• Type: Conference
• Year: 2008
• Where: NIPS
• Authors: Yik-Cheung Tam, Tanja Schultz
• Keywords: Bigram Lsa, Bigram Lsa Approach, Information Technology, NIPS 2008, Unigram Lsa