Investigation of Practical Aspects of Single Channel Speech Separation for ASR

TL;DR

This paper enhances single channel speech separation for ASR by combining a two-stage training scheme with model compression, leading to significant WER improvements on LibriCSS with lightweight models.

Contribution

It introduces a novel two-stage training approach and a modified teacher-student technique for model compression in speech separation for ASR.

Findings

Achieved 2.70% absolute WER reduction on LibriCSS.

Developed a lightweight model with less than 10M parameters.

Demonstrated improved performance in both utterance-wise and continuous evaluation.

Abstract

Speech separation has been successfully applied as a frontend processing module of conversation transcription systems thanks to its ability to handle overlapped speech and its flexibility to combine with downstream tasks such as automatic speech recognition (ASR). However, a speech separation model often introduces target speech distortion, resulting in a sub-optimum word error rate (WER). In this paper, we describe our efforts to improve the performance of a single channel speech separation system. Specifically, we investigate a two-stage training scheme that firstly applies a feature level optimization criterion for pretraining, followed by an ASR-oriented optimization criterion using an end-to-end (E2E) speech recognition model. Meanwhile, to keep the model light-weight, we introduce a modified teacher-student learning technique for model compression. By combining those approaches, we achieve a absolute average WER improvement of 2.70% and 0.77% using models with less than 10M parameters compared with the previous state-of-the-art results on the LibriCSS dataset for utterance-wise evaluation and continuous evaluation, respectively