Acoustic Factorisation for Robust Speech Recognition

Add to your list(s) Download to your calendar using vCal

• Eric(Yongqiang) Wang (University of Cambridge)
• Friday 11 January 2013, 12:30-13:30
• Department of Engineering - LR6.

If you have a question about this talk, please contact Catherine Breslin.

For many practical scenario, speech recognition systems need to be robust against multiple acoustic factors, e.g., speaker and noise differences. A conventional approach would be adapting acoustic models by transforms estimated for each speaker and environment combination. However, an ideal approach would be based on the concept of acoustic factorisation, where transforms are factorised such that each component transform only models one distinct acoustic factor. This gives flexibility for model adaptation, e.g., rapid speaker adaptation in fast changing environments, as demonstrated by the experiments. There are a few options to construct factorised transform: the component transforms can be constrained to have different forms such that different distortions can be modelled separately and/or by imposing overlapping data such that the component transforms learn different attributes. These options as well as future works will be discussed in the talk.

This talk is part of the CUED Speech Group Seminars series.

This talk is included in these lists:

• CUED Speech Group Seminars
• Cambridge Forum of Science and Humanities
• Cambridge Language Sciences
• Cambridge talks
• Chris Davis' list
• Department of Engineering - LR6
• Guy Emerson's list
• Information Engineering Division seminar list
• PhD related

Note that ex-directory lists are not shown.