On Architectures and Training for Raw Waveform Feature Extraction in ASR

TL;DR

This paper evaluates the effectiveness of the wav2vec raw waveform feature extractor in hybrid ASR systems without extra untranscribed data, comparing it to traditional and supervised features, and explores combining features for improved performance.

Contribution

It systematically assesses wav2vec in a data-limited setting and investigates feature combination strategies to enhance ASR accuracy.

Findings

Wav2vec features outperform standard Gammatone features in ASR.

Combining wav2vec with supervised features yields further improvements.

Pre-trained wav2vec benefits are evident even without additional untranscribed data.

Abstract

With the success of neural network based modeling in automatic speech recognition (ASR), many studies investigated acoustic modeling and learning of feature extractors directly based on the raw waveform. Recently, one line of research has focused on unsupervised pre-training of feature extractors on audio-only data to improve downstream ASR performance. In this work, we investigate the usefulness of one of these front-end frameworks, namely wav2vec, in a setting without additional untranscribed data for hybrid ASR systems. We compare this framework both to the manually defined standard Gammatone feature set, as well as to features extracted as part of the acoustic model of an ASR system trained supervised. We study the benefits of using the pre-trained feature extractor and explore how to additionally exploit an existing acoustic model trained with different features. Finally, we systematically examine combinations of the described features in order to further advance the performance.