Towards End-to-end Unsupervised Speech Recognition

TL;DR

This paper introduces wav2vec-U 2.0, an end-to-end unsupervised speech recognition model that eliminates pre-processing steps and enhances accuracy with a new architecture and self-supervised objectives, improving results across languages.

Contribution

It presents wav2vec-U 2.0, a novel end-to-end unsupervised speech recognition framework that simplifies the pipeline and boosts performance using architectural improvements and auxiliary objectives.

Findings

Improved unsupervised recognition accuracy across multiple languages.

Elimination of audio pre-processing steps.

Simpler, more effective architecture.

Abstract

Unsupervised speech recognition has shown great potential to make Automatic Speech Recognition (ASR) systems accessible to every language. However, existing methods still heavily rely on hand-crafted pre-processing. Similar to the trend of making supervised speech recognition end-to-end, we introduce wav2vec-U 2.0 which does away with all audio-side pre-processing and improves accuracy through better architecture. In addition, we introduce an auxiliary self-supervised objective that ties model predictions back to the input. Experiments show that wav2vec-U 2.0 improves unsupervised recognition results across different languages while being conceptually simpler.