Combined static and dynamic variance adaptation for efficient interconnection of speech enhancement pre-processor with speech recognizer

摘要: It is well known that automatic speech recognition performs poorly in presence of noise or reverberation. Much research has been undertaken on model adaptation and enhancement to increase the robustness recognizers. Model effective remove static mismatch between features acoustic parameters, but may not cope with dynamic mismatch. Speech approaches can reduce perturbations, often do interconnect recognizer. There seems be a lack optimal way combine these two approaches. In this paper we propose introducing capabilities into scheme. We focus variance adaptation, novel parametric includes components. The component derived from pre-process, parameters are optimized using an adaptive training An evaluation method dereverberation for preprocessing revealed 80 % relative error rate reduction was possible compared dereverberated speech, final 5.4 which close clean (1.2%).