More about HKUST
The Use of Discrete Hidden Markov Model with a Very Large Codebook for Automatic Speech Recognition
PhD Thesis Proposal Defence Title: "The Use of Discrete Hidden Markov Model with a Very Large Codebook for Automatic Speech Recognition" by Mr. Guoli Ye ABSTRACT: With the advance of semiconductor technology and the popularity of distributed automatic speech recognition (ASR) paradigm (e.g., Siri in iPhone4s), we would like to revisit the discrete hidden Markov model (DHMM) as the acoustic model in ASR. Compared with continuous density hidden Markov model (CDHMM), the dominant acoustic model used in modern ASR systems, DHMM has inherently attractive properties: it uses non-parametric state output distributions and takes only O(1) time to get the probability value from it; Furthermore, the discrete features used in DHMM, compared with cepstral coefficients in CDHMM, could be encoded in fewer bits, lowering the bandwidth requirement in distributed speech recognition architecture. Unfortunately, the recognition performance of conventional DHMMis significantly worse than that of CDHMM due to the large quantization error and the use of multiple independent streams. In this proposal, we propose to reduce the quantization error of DHMM by using a very large codebook with tens of thousands of codewords (in conventional DHMM, the number of codewords in a codebook usually ranges from 256 to 1024). An extensive literature review is given in the proposal to show that very large codebook in DHMMis novel and necessary. The challenges to use large codebook are discussed, with a novel framework called subspace high-density discrete HMM (SHDDHMM) to solve the problems. A large vocabulary continuous speech recognition task is used to evaluate the proposed framework, showing the feasibility and benefits of DHMM with very large codebook. Future works for completion of the thesis are also given in the proposal. Date: Thursday, 7 June 2012 Time: 2:00pm - 4:00pm Venue: Room 3315 lifts 17/18 Committee Members: Dr. Brian Mak (Supervisor) Prof. James Kwok (Chairperson) Dr. Raymond Wong Prof. Dit-Yan Yeung **** ALL are Welcome ****