This is the honours Speech Recognition Unit. The Unit Notes contain all of the lecture details etc. The class Wiki is used for collaborative work on training the Sphinx Speech Recogniser.

Essay Topics

The following are the essay topics up for grabs. The intention is that you write a review of one or two papers in the area you choose. Write around 2000 words, or enough to convey the core of the ideas in the paper(s).

1. Speaker Identification/Verification: a detailed look at one or two SV systems or an evaluation of SV applications in terms of the false accept/false reject tradeoff and the kinds of systems that are used in various applications.
2. Robust Speech Recognition: Recognition in noisy environments, the different signal processing techniques used to compensate for noise in, say, a car.
3. Confidence Measures: modifications to the HMM algorithms to provide a measure of confidence in the results that are returned.
4. Neural Networks in ASR: There are a number of systems which use hybrid HMM/Neural network recognisers, discuss the role of the NN in one of these architectures and look at the relative performance of these systems and standard HMMs.
5. Language Modelling: A detailed look at stochastic language models and alternatives.
6. State of the Art in ASR: Review the results of the NIST broadcast news trials, summarise the performance of the current systems and briefly describe some of the things that these systems do beyond what we've discussed in the basic HMM recogniser.

Please send any comments on these pages to Steve Cassidy .
Copyright (c) 2002 by Macquarie University. All rights reserved.