Filterbank design for end-to-end speech separation
This work addresses speech separation for noisy audio processing, offering incremental improvements in filterbank design.
The authors tackled the problem of single-channel speech separation by extending real-valued learned and parameterized filterbanks into complex-valued analytic versions, evaluating them on the WHAM dataset. The results showed that the proposed analytic learned filterbank consistently outperformed ConvTasNet's real-valued filterbank, with complex-valued representations and masks being beneficial in all conditions.
Single-channel speech separation has recently made great progress thanks to learned filterbanks as used in ConvTasNet. In parallel, parameterized filterbanks have been proposed for speaker recognition where only center frequencies and bandwidths are learned. In this work, we extend real-valued learned and parameterized filterbanks into complex-valued analytic filterbanks and define a set of corresponding representations and masking strategies. We evaluate these filterbanks on a newly released noisy speech separation dataset (WHAM). The results show that the proposed analytic learned filterbank consistently outperforms the real-valued filterbank of ConvTasNet. Also, we validate the use of parameterized filterbanks and show that complex-valued representations and masks are beneficial in all conditions. Finally, we show that the STFT achieves its best performance for 2ms windows.