KL-Divergence Guided Two-Beam Viterbi Algorithm on Factorial HMMs

Yeh, Raymond

Permalink

https://hdl.handle.net/2142/55632 Copy

Description

Title

KL-Divergence Guided Two-Beam Viterbi Algorithm on Factorial HMMs

Author(s)

Yeh, Raymond

Contributor(s)

Hasegawa-Johnson, Mark

Issue Date

2014-05

Keyword(s)

• factorial hidden Markov model
• Viterbi beam
• digit recognition

Abstract

This thesis addresses the problem of the high computation complexity issue that arises when decoding hidden Markov models (HMMs) with a large number of states. A novel approach, the two-beam Viterbi, with an extra forward beam, for decoding HMMs is implemented on a system that uses factorial HMM to simultaneously recognize a pair of isolated digits on one audio channel. The two-beam Viterbi algorithm uses KL-divergence and hierarchical clustering to reduce the overall decoding complexity. This novel approach achieves 60% less computation compared to the baseline algorithm, the Viterbi beam search, while maintaining 82.5% recognition accuracy.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/55632

Owning Collections