Accelerating lattice scoring of automatic speech recognition through acoustic pre-pruning on GPU

This thesis introduces an acoustic pre-pruning algorithm that speeds up lattice scoring for GMM based ASR systems, and a constrained agglomerative clustering algorithm that makes it possible to maintain the advantage of the new algorithm in a GPU implementation. The implementation undergoes 2% to 6% degradation in PER while accelerating the runtime of lattice scoring by 45X to 60X over a traditional CPU implementation.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• He, Di

Contributors dc:contributor

• Chen, Deming

• acoustic pre-pruning
• lattice scoring
• Gaussian mixture model (GMM)
• graphic processing unit (GPU)

Statement dc:rights

• Copyright 2014 Di He

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/73019

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/73019