Published In Mixture Pruning and Roughening for Scalable Acoustic Models

In an automatic speech recognition system us-ing a tied-mixture acoustic model, the main cost in CPU time and memory lies not in the evaluation and storage of Gaussians them-selves but rather in evaluating the mixture likelihoods for each state output distribution. Using a simple entropy-based technique for pruning the mixture weight distributions, we can achieve a significant speedup in recogni-tion for a 5000-word vocabulary with a negli-gible increase in word error rate. This allows us to achieve real-time connected-word dicta-tion on an ARM-based mobile device.