Language modeling for robust balancing of acoustic and linguistic probabilities
Atsunori Ogawa, Kazuya Takeda, Fumitada Itakura · 2002
The length of a word sequence is not taken into account under language modeling in n-gram local probability modeling. Due to this property, the optimal value of the language weight for balancing acoustic and linguistic probabilities is affected by the sequence length. To deal with this problem, a new language model is developed based on the Bernoulli trial model. By taking the sequence length into account, not only is better recognition accuracy achieved, but also more robust balancing with the acoustic probability, as compared with the normal n-gram model.