In-car speech recognition based on multiple regression of spectra

摘要

This paper describes a new multichannel method of noisy speech recognition, which estimates the log spectrum of speech at a close-talking microphone based on the multiple regression of the log spectra (MRLS) of noisy signals captured by the distributed microphones. Since the method does not assume the arrangement of sound sources and microphones, it can be applied to in-car speech recognition directly. The experimental evaluation shows an error reduction of up to 15% in isolated word recognition accuracy under various driving conditions.