Out There - Speech recognition in the far field

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

• Thomas Hain, Professor of Speech and Audio Technology, University of Sheffield
• Thursday 27 November 2014, 14:00-15:00
• Gisbert Kapp, N123.

If you have a question about this talk, please contact Dr. Philip Weber.

Note unusual time

Speech recognition from far field microphones has always been a niche interest, mostly supported by researchers with strong interest in signal processing. Standard recognition techniques have seen steady progress over the last decade, and with the re-emergence of neural networks, speech recognition can now be used in much more natural situations. Recent consumer products clearly highlight a renewed interest in far field processing, which is also underlined by an iARPA challenge. Until recently speech enhancement based approaches have dominated the scene, however new results show that these can be outperformed with relatively simple machine learning techniques. In this talk I will present our ongoing work on far field recognition using multiple microphones, based on the concept of environment models. Here the objective is not to deconstruct the signals into their components, but to inform models on context and benefit from targeted optimisation. I will present work on maximum likelihood beamforming, deep neural network frontends for far field recognition, and acoustic factorisation.

This talk is part of the Speech Recognition by Synthesis Seminars series.

This talk is included in these lists:

• Gisbert Kapp, N123
• Speech Recognition by Synthesis Seminars

Note that ex-directory lists are not shown.