Robust Speech Recognition via Large-Scale Weak Supervision
This provides a foundation for robust speech processing by enabling zero-shot transfer across tasks, though it builds incrementally on large-scale weak supervision approaches.
The researchers tackled speech recognition by training models on 680,000 hours of internet audio transcripts, achieving competitive results with prior supervised methods in a zero-shot setting without fine-tuning, and approaching human-level accuracy and robustness.
We study the capabilities of speech processing systems trained simply to predict large amounts of transcripts of audio on the internet. When scaled to 680,000 hours of multilingual and multitask supervision, the resulting models generalize well to standard benchmarks and are often competitive with prior fully supervised results but in a zero-shot transfer setting without the need for any fine-tuning. When compared to humans, the models approach their accuracy and robustness. We are releasing models and inference code to serve as a foundation for further work on robust speech processing.