Disfluency Detection with Unlabeled Data and Small BERT Models
This addresses the need for efficient, local disfluency detection models in automatic speech recognition pipelines, but it is incremental as it builds on existing BERT and data augmentation methods.
The paper tackled the problem of reducing model size and inference time for disfluency detection to enable on-device use, achieving high performance with models as small as 1.3 MiB.
Disfluency detection models now approach high accuracy on English text. However, little exploration has been done in improving the size and inference time of the model. At the same time, automatic speech recognition (ASR) models are moving from server-side inference to local, on-device inference. Supporting models in the transcription pipeline (like disfluency detection) must follow suit. In this work we concentrate on the disfluency detection task, focusing on small, fast, on-device models based on the BERT architecture. We demonstrate it is possible to train disfluency detection models as small as 1.3 MiB, while retaining high performance. We build on previous work that showed the benefit of data augmentation approaches such as self-training. Then, we evaluate the effect of domain mismatch between conversational and written text on model performance. We find that domain adaptation and data augmentation strategies have a more pronounced effect on these smaller models, as compared to conventional BERT models.