Robust Speech Recognition via Large-Scale Weak Supervision

Abstract

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 competitive with prior fully supervised results without the need for any fine-tuning.

Key Findings

• 1Trained on 680,000 hours of multilingual audio data from the internet
• 2Achieved near-human-level speech recognition without task-specific fine-tuning
• 3Supported transcription in 97 languages and translation to English
• 4Demonstrated robust performance across accents, noise, and domains
• 5Released as open-source enabling broad adoption

Impact & Significance