Continuous Speech Separation with Conformer
This work addresses speech separation for conversation transcription, showing incremental improvements over existing methods.
The paper tackled the problem of continuous speech separation by replacing recurrent neural networks with transformer and conformer models to capture global information, achieving state-of-the-art results with a 23.5% relative WER reduction in utterance-wise evaluation and 15.4% in continuous evaluation on the LibriCSS dataset.
Continuous speech separation plays a vital role in complicated speech related tasks such as conversation transcription. The separation model extracts a single speaker signal from a mixed speech. In this paper, we use transformer and conformer in lieu of recurrent neural networks in the separation system, as we believe capturing global information with the self-attention based method is crucial for the speech separation. Evaluating on the LibriCSS dataset, the conformer separation model achieves state of the art results, with a relative 23.5% word error rate (WER) reduction from bi-directional LSTM (BLSTM) in the utterance-wise evaluation and a 15.4% WER reduction in the continuous evaluation.