ABSTRACT
This chapter presents a survey of speech processing research, mainly related to automatic speech recognition (ASR), carried out during the IM2 NCCR. Section 15.1 briefly introduces the two most popular pattern matching approaches to ASR, namely, hidden Markov model (HMM) based speech recognition (a stochastic pattern matching approach) and instance/template based speech recognition (a deterministic pattern matching approach). In terms of speech acquisition and front-end processing, IM2 NCCR presented new challenges such as how to process multiple microphone speech, how to localize multiple speakers, and how to achieve noise robustness in single-channel close-talking speech and multi-channel distant speech. Section 15.2 presents briefly the advancements in these directions.
