Towards End-to-end Unsupervised Speech Recognition
This work addresses making ASR systems accessible to every language by simplifying unsupervised methods, though it appears incremental as it builds on prior wav2vec-U approaches.
The paper tackles the problem of unsupervised speech recognition by introducing wav2vec-U 2.0, which eliminates audio-side pre-processing and improves accuracy through better architecture and an auxiliary self-supervised objective, resulting in improved recognition results across different languages.
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.