Robust Speech Recognition via Large-Scale Weak Supervision (Whisper)

Trained an encoder-decoder transformer on 680,000 hours of multilingual, multitask supervised audio from the web, achieving human-level robustness across noise and accents without dataset fine-tuning.

Directly guides architectural decisions, alignment strategy, and serving infrastructure for task-speech-recognition, task-machine-translation.

Assumptions

• Empirical distribution regularity holds and target domain adheres to pretraining linguistic/visual support

Limitations

• Resource scaling, inference memory requirements, and alignment robustness vary with model size and hardware topology