University of Cambridge > Talks.cam > CUED Speech Group Seminars > Robust speech recognition / Towards better probabilistic models of speech - a discussion

Robust speech recognition / Towards better probabilistic models of speech - a discussion

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

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

BRIDGING THE GAP BETWEEN SPEECH ENHANCEMENT AND RECOGNITION , Takuya Yoshioka

One major goal of ASR is to accurately transcribe target utterances even in acoustically adverse environments and this has been tackled with a range of approaches. The approaches to noise robust ASR systems can be broadly classified into two categories. One is based on speech enhancement, which attempts to clean speech signals that are corrupted by noise. The speech enhancement approaches are usually optimized using SNR or related critetia. The approaches in the other category, including feature/model-space VTS , JUD, and SPLICE , are more tailored to ASR in the sense that they exploit statistical models of clean feature vectors for effective compensation for noise. In this short talk, I will argue that these two types of approaches have different advantages and present a novel general scheme for integrating them. I will also present applications of this scheme to meeting transcription and reverberant speech recognition. I try to explain the basic concept without using too many equations so that the talk would be enjoyable for everyone in this group.

TOWARDS BETTER PROBABILISTIC MODELS OF SPEECH : WHY SAMPLED TRAJECTORIES SOUND BAD AND HOW TO FIX THEM - A DISCUSSION , Matt Shannon

Matt will present some preliminary experimental results and some thought provoking speech samples for discussion.

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

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2020 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity