A comparison of handcrafted, parameterized, and learnable features for speech separation

The design of acoustic features is important for speech separation. It can be roughly categorized into three classes: handcrafted, parameterized, and learnable features. Among them, learnable features, which are trained with separation networks jointly in an end-to-end fashion, become a new trend of modern speech separation research, e.g., convolutional time domain audio separation network (Conv-Tasnet), Dual-path RNN(DPRNN), Dual-path Transformer Network(DPTNet). At the same time, handcrafted and parameterized features are also shown competitively in very recent studies. However, a systematic comparison across the three kinds of acoustic features has not been conducted yet. This paper compares them in the framework of Conv-Tasnet, DPRNN, and DPTNet by setting its encoder and decoder with different acoustic features. We also generalize the handcrafted multi-phase gammatone filterbank (MPGTF) to a new parameterized multi-phase gammatone filterbank (ParaMPG TF). Experimental results on the WSJ0-2mix corpus and WHAMR! corpus show that if the decoder is learnable, then setting the encoder to STFT, MPGTF, ParaMPGTF, and learnable features lead to similar performance on Conv- Tasnet, DPRNN, and DPTNet; and when the pseudo-inverse transforms of STFT, MPGTF, and ParaMPGTF are used as the decoders on Conv- Tasnet, the proposed ParaMPGTF performs better than the other two handcrafted features.