This paper: Implementing Parallel Speech Decoding in HTK Toolkit by Exploiting Manycore GPU and SIMD Technology discusses parallel speech decoding using HTK using the NVIDIA CUDA platform.

NSH's blog Speech Recognition With CMU Sphinx also points to papers (cuHMM: a CUDA Implementation of Hidden Markov Model
Training and Classication) discussing 200 x improvement on NVidia CUDA.

From Wikipedia:

CUDA (an acronym for Compute Unified Device Architecture) is a parallel computing architecture developed by NVIDIA. CUDA is the computing engine in NVIDIA graphics processing units or GPUs that is accessible to software developers through industry standard programming languages.

Basically you use your Nvidia, CUDA-enabled graphics card to do the heavy lifting in acoustic model training and/or for speech recognition, resulting in big performance improvements.

--- (Edited on 12/18/2009 1:36 pm [GMT-0500] by kmaclean) ---