Potential projects for visitors, new group members, and EE225D students

For brief descriptions of ongoing ICSI Speech Group research projects, see here. For reports of recent research work in the ICSI Speech Group, see the ICSI publications listings. For a few descriptions of graduate student research topics in the ICSI Speech Group, see here.

The following is a very incomplete list of potential projects related to the research interests of the ICSI Speech Group. It is very incomplete because most of these suggestions were written by David Gelbart so they relate to his research interests in particular.

• Amplitude filtering for noise-robust speech recognition
• Unsupervised MLP adaptation
• Possible projects on speech recognition without a close microphone
• Multi-stream back-propagation for neural network training
• u-law encoding for meeting recognition training
• Bandwidth extension for training meeting recognition
• Look for floating point accuracy problems in speech processing software
• Compare cepstral mean subtraction and log-DFT mean normalization

225D students who are not members of the ICSI Speech Group should note that some projects are easier to do independently than others. In particular, projects involving the recognition of conversational speech will probably require a significant amount of help from a member of the Speech Group. Before committing to such a project, you should check whether such help will be available.

225D students may also be interested in project ideas from various open source audio projects outside ICSI. The Summer of Code web site includes ideas from the Xiph project and other audio-related projects. The VoxForge project has a list of ideas here. The Sourceforge Help Wanted listings often include audio projects.

For a list of some speech-processing-related online tutorials, and online directories of tools and databases, click here.