Speech model · Established · Intermediate

Whisper

An open speech recognition model family trained on large multilingual audio, widely used for transcription and translation.

What Whisper is

Whisper transcribes speech across many languages and can translate into English, with robustness to accents and background noise that made it a default open choice.

How it works

Weights are published and can be run locally in several optimised implementations, or accessed through hosted APIs. Smaller variants run in real time on modest hardware.

Why it matters

It made high-quality transcription free and self-hostable, which changed the economics of captioning and meeting tooling.

Common uses

→Meeting and podcast transcription
→Subtitling
→Voice interfaces
→Call analytics

Strengths

✓Open weights
✓Strong multilingual coverage
✓Runs locally

Watch for

✓Can hallucinate text in silence or noise
✓Diarisation needs a separate model

Continue exploring

Speech Recognition OpenAI (2)Open-Weight Model