Detection of overlapped speech using lapel microphones in meeting

Abstract

We propose an overlapped speech detection method for speech recognition and speaker diarization of meetings, where each speaker wears a lapel microphone. Two novel features are utilized as inputs for a GMM-based detector. One is speech power after cross-channel spectral subtraction which reduces the power from the other speakers. The other is an amplitude spectral cosine correlation coefficient which effectively extracts the correlation of spectral components in a rather quiet condition. We evaluated our method using a meeting speech corpus of four speakers. The accuracy of our proposed method, 75.7%, was significantly better than that of the conventional method, 66.8%, which uses raw speech power and power spectral Pearson's correlation coefficient.

Multi-Talker Speech Recognition Based on Blind Source Separation with ad hoc Microphone Array Using Smartphones and Cloud Storage

“…In ASR in a multi-talker environment, the overlap of speech sounds causes problems [4]. The detection of speech overlaps was studied to avoid degrading the performance of ASR [5]. In the Pascal Speech Separation Challenge [6], recognizing a target speech in the presence of another talker's speech was evaluated in a monaural scenario.…”

Analysis and development of robust speaker diarization for meetings

“…Overlap in meetings recorded using lapel microphones is investigated in (Yokoyama, Nasu, Iwano, & Shinoda, 2013). In this work two novel features are presented: speech power after cross-channel spectral subtraction and the so called "amplitude spectral cosine correlation coefficient" which calculates the cosine correlation between the spectrum of two microphones.…”