Whisper
Speech recognition model trained on large-scale weakly supervised audio.
Overview
Whisper is a speech recognition model trained on a very large, varied set of audio with weak labels, which makes it robust across accents, background noise and recording quality. It solves the problem that earlier systems were tuned to clean datasets and degraded sharply outside them. It exists because scale and diversity of audio mattered more than a bespoke acoustic model, and it is the default open option for transcription.
Technology Graph
How Whisper connects to the rest of the stack.
Uses and depends on
- Transformer architecture Algorithms Whisper based on Transformer High ✓ source
Applied to
- Speech Recognition application Applications Whisper used for Speech Recognition High ✓ source