Digital Signal Processing Reference
In-Depth Information
Chapter 19
IN-CAR SPEECH RECOGNITION USING
DISTRIBUTED MICROPHONES
Tetsuya Shinde
1
, Kazuya Takeda
2
, Fumitada Itakura
1
1 2
Graduate School of Engineering; Graduate School of Information Science Nagoya Uni-
versity, 1 Furo-cho, Nagoya 464-8603 Japan
Email: takeda@is.nagoya-u.ac.jp
Abstract
In this paper, we describe a method for multichannel noisy speech recogni-
tion that can adapt to various in-car noise situations during driving. Our
proposed technique enables us to estimate the log spectrum of speech at a
close-talking microphone based on the multiple regression of the log spec-
tra (MRLS) of noisy signals captured by a set of distributed microphones.
Through clustering of the spatial noise distributions under various driv-
ing conditions, the regression weights for MRLS are effectively adapted
to the driving conditions. The experimental evaluation shows an aver-
age error rate reduction of 43 % in isolated word recognition under 15
different driving conditions.
Keywords:
In-car-ASR, multiple microphone, linear regression
1. INTRODUCTION
Array-microphone signal processing is known for sometime now to be
effective for spatially selective signal capture and, in particular, noisy
speech recognition when the locations of the speaker and noise sources are
predetermined. However, when the spatial configuration of the speaker
and noise sources are unknown or they change continuously, it is not easy
to steer the directivity adaptively to the new conditions [1], [2], [3].
Previously, we have proposed multiple regression of log spectra (MRLS)
to improve the robustness in the case of a small perturbation of the spa-
tial distribution of the source and noise signals. In that study, log spectra
