MULTI-DECODER DPRNN: HIGH ACCURACY SOURCE COUNTING AND SEPARATION

Content Files

FA20-ECE499-Thesis-Zhu, Junzhe.pdf

Permalink

https://hdl.handle.net/2142/125114

Title: MULTI-DECODER DPRNN: HIGH ACCURACY SOURCE COUNTING AND SEPARATION
Author(s): Zhu, Junzhe
Issue Date: 2020-12-01
Keyword(s): Source separation
Date of Ingest: 2024-11-14T10:08:40-06:00
Abstract: We propose an end-to-end trainable approach to single-channel speech separation with unknown number of speakers. Our approach extends the MulCat source separation backbone with additional output heads: a count-head to infer the number of speakers, and decoder-heads for reconstructing the original signals. Beyond the model, we also propose a metric on how to evaluate source separation with variable number of speakers. Specifically, we cleared up the issue on how to evaluate the quality when the ground-truth has more or less speakers than the ones predicted by the model. We evaluate our approach on the WSJ0-mix datasets, with mixtures up to five speakers. We demonstrate that our approach outperforms stateof-the-art in counting the number of speakers and remains competitive in quality of reconstructed signals.
Type of Resource: text
Genre of Resource: dissertation/thesis
Language: eng

The best of ECE undergraduate research