Tackling the Acoustic Front-end for Distant-Talking Automatic Speech Recognition
From HLT@INESC-ID
Date
- 15:30, Tuesday, October 02, 2007
- Room 336
Speaker
- Walter Kellermann, Erlangen-Nuremberg University, Distinguished Lecturer of the IEEE Signal Processing Society.
Abstract
With the ever-growing interest in 'natural' hands-free acoustic human/machine interfaces, the need for according distant-talking automatic speech recognition (ASR) systems increases.
Considering interactive TV as a challenging exemplary application scenario, we investigate the structural problems presented by noisy and reverberant multi-source environments with unpredictable interference and acoustic echoes of loudspeaker signals, and discuss current acoustic signal processing techniques to enhance the input to the actual ASR system. Special attention is paid to reverberation, which affects speech recognizers much more than human listeners, and a recently published method incorporating a reverberation model on the feature level of ASR is discussed.