Techniques for Noise Robustness in Automatic Speech Recognition

10

Reverberant Speech Recognition

Reinhold Haeb-Umbach, Alexander Krueger

University of Paderborn, Germany

10.1 Introduction

From a usage point of view, there are a number of reasons why in many applications of automatic speech recognition (ASR) distant talking microphones are to be preferred over close-talking microphones. The first is convenience: freeing the user from holding a microphone or wearing a headset increases the ease of use, and thus raises the acceptance of appliances or services operated by voice commands. A second reason is safety: there are numerous applications, where the hands are needed for more important tasks than for holding a microphone to capture the user's speech. Examples include the hands-free control of a cellular phone or a car navigation system while driving, or the control of some apparatus by a surgeon while being busy with an operation. Finally, moving the microphone away from the mouth of the speaker is in line with the disappearing computer and the ambient intelligence paradigm, which has been put forward already for several years [1]. It describes the vision of technology that is invisible, embedded in our surroundings while still being present whenever we need it. Interacting with it should be simple and effortless, and speech, as a “remote control” that a user has with him all the time, is the ideal means of interaction.

However, increasing the distance between the speaker and the microphone has dramatic consequences on the quality of the ...