An Unsupervised Approach to Cochannel Speech Separation

Abstract

Abstract-Cochannel (two-talker) speech separation is predominantly addressed using pretrained speaker dependent models. In this paper, we propose an unsupervised approach to separating cochannel speech. Our approach follows the two main stages of computational auditory scene analysis: segmentation and grouping. For voiced speech segregation, the proposed system utilizes a tandem algorithm for simultaneous grouping and then unsupervised clustering for sequential grouping. The clustering is performed by a search to maximize the ratio of between-and within-group speaker distances while penalizing within-group concurrent pitches. To segregate unvoiced speech, we first produce unvoiced speech segments based on onset/offset analysis. The segments are grouped using the complementary binary masks of segregated voiced speech. Despite its simplicity, our approach produces significant SNR improvements across a range of input SNR. The proposed system yields competitive performance in comparison to other speaker-independent and model-based methods.

Speaker-dependent multipitch tracking using deep neural networks

“…In addition to the above comparisons, we have compared with Hu and Wang (2013), where a clustering algorithm is used to group short pitch contours into two speakers. We found that this method performs better than Jin and Wang's method, but worse than the speaker-dependent method by Wohlmayr et al The details of this comparison can be found in Liu and Wang (2015b).…”

“…Although most speaker-independent models perform well for estimating pitch periods, they can not assign pitch estimates to the underlying speakers for multipitch tracking. To alleviate this problem, Hu and Wang (2013) build their system on the tandem algorithm (Hu and Wang, 2010) and group simultaneous pitch contours into two speakers using a constrained clustering algorithm. Similarly, Duan et al (2014) take the pitch estimates of speaker-independent multipitch trackers as input and stream pitch points by clustering.…”