Optimization of Speaker-aware Multichannel Speech Extraction with ASR Criterion

conference paper

English

This paper addresses the problem of recognizing speech corrupted by overlapping speakers in a multichannel setting. To extract a target speaker from the mixture, we use a neural network based beamformer which uses masks estimated by a neural network to compute statistically optimal spatial filters. Following our previous work, we inform the neural network about the target speaker using information extracted from an adaptation utterance, enabling the network to track the target speaker. While in the previous work, this method was used to separately extract the speaker and then pass such preprocessed speech to a speech recognition system, here we explore training both systems jointly with a common speech recognition criterion. We show that integrating the two systems and training for the final objective improves the performance. In addition, the integration enables further sharing of information between the acoustic model and the speaker extraction system, by making use of the predicted HMMstate posteriors to refine the masks used for beamforming.

Speaker extraction, joint training, speaker adaptive neural network, beamforming, speech recognition

ŽMOLÍKOVÁ, K.; DELCROIX, M.; KINOSHITA, K.; HIGUCHI, T.; NAKATANI, T.; ČERNOCKÝ, J.

15. 4. 2018

IEEE Signal Processing Society

Calgary

978-1-5386-4658-8

Proceedings of ICASSP 2018

6702

6706

5

https://www.fit.vut.cz/research/publication/11722/

zmolikova_icassp2018_0006702.pdf