Multi-Channel Speaker Verification with Conv-Tasnet Based Beamformer

článek ve sborníku ve WoS nebo Scopus

angličtina

We focus on the problem of speaker recognition in far-field multichannel data. The main contribution is introducing an alternative way of predicting spatial covariance matrices (SCMs) for a beamformer from the time domain signal. We propose to use ConvTasNet, a well-known source separation model, and we adapt it to perform speech enhancement by forcing it to separate speech and additive noise. We experiment with using the STFT of Conv-TasNet outputs to obtain SCMs of speech and noise, and finally, we fine-tune this multi-channel frontend w.r.t. speaker verification objective. We successfully tackle the problem of the lack of a realistic multichannel training set by using simulated data of MultiSV corpus. The analysis is performed on its retransmitted and simulated test parts. We achieve consistent improvements with a 2.7 times smaller model than the baseline based on a scheme with mask estimating NN.

Conv-TasNet, beamforming, embedding extractor, speaker verification, MultiSV

MOŠNER, L.; PLCHOT, O.; BURGET, L.; ČERNOCKÝ, J.

27. 5. 2022

IEEE Signal Processing Society

Singapore

978-1-6654-0540-9

ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings

7982

7986

5

https://ieeexplore.ieee.org/document/9747771

mosner_icassp2022_Multi-Channel_Speaker_Verification_with_Conv-Tasnet_Based_Beamformer.pdf