# Investigation of Practical Aspects of Single Channel Speech Separation for ASR

arXiv: https://arxiv.org/abs/2107.01922 · HTML: https://mlsanity.com/paper/2107.01922/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 5 Jul 2021.

## Built on by (most cited first, 8 of 8)

- [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://mlsanity.com/paper/2110.13900/) (2021)
- [Streaming Multi-Talker ASR with Token-Level Serialized Output Training](https://mlsanity.com/paper/2202.00842/) (2022)
- [Speech separation with large-scale self-supervised learning](https://mlsanity.com/paper/2211.05172/) (2022)
- [Serialized Output Training by Learned Dominance](https://mlsanity.com/paper/2407.03966/) (2024)
- [Leveraging Real Conversational Data for Multi-Channel Continuous Speech Separation](https://mlsanity.com/paper/2204.03232/) (2022)
- [Handling Trade-Offs in Speech Separation with Sparsely-Gated Mixture of Experts](https://mlsanity.com/paper/2211.06493/) (2022)
- [A Glance is Enough: Extract Target Sentence By Looking at A keyword](https://mlsanity.com/paper/2310.05352/) (2023)
- [Joint Beamforming and Speaker-Attributed ASR for Real Distant-Microphone Meeting Transcription](https://mlsanity.com/paper/2410.21849/) (2024)

## Abstract (arXiv)

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