Hidden Markov Model for Isolated Word Recognition on GPU
Author: Leiming Yu
ylm@ece.neu.edu
Northeastern University, Boston, MA, USA
It is implemented in the following paper. Please refer to it if you want to use. Thanks.
@inproceedings{yu2014gpu,
title={GPU-accelerated HMM for Speech Recognition},
author={Yu, Leiming and Ukidave, Yash and Kaeli, David},
booktitle={Parallel Processing Workshops (ICCPW), 2014 43rd International Conference on},
pages={395--402},
year={2014},
organization={IEEE}
}
Usage:
To compile:
$cmake .
$make
To verify each module:
$make test
or run each test separately:
$./test_forword
$./test_backwd
$./test_bw
To run program:
$./hmmsr number_of_hidden_states number_of_streams
e.g., $./hmmsr 256 4
In CMakeLists.txt, use NVCC compiler flags to specify which feature to use.
-DHQ=1 (turn on hyper-q)
-DHQ=0 (turn off hyper-q)