Binaural processing for robust recognition of degraded speech

摘要: This paper discusses a new combination of techniques that help in improving the accuracy speech recognition adverse conditions using two microphones. Classic approaches toward binaural processing use some form cross-correlation over time across sensors to effectively isolate target from interferers. Several additional temporal and spatial masking have been proposed past improve presence reverberation interfering talkers. In this paper, we consider frequency limited range channels addition existing methods monaural processing. has effect locating reinforcing coincident peaks representation interaction provides local smoothing specified frequencies. Combined with mentioned above, leads significant improvements recognition.