Research-only transcription models
8 in the catalogue today, and the list grows as the market does. Every one with what it costs, who sells it and where it stands.
Models that turn recorded speech into text. They are metered by the minute or the second of audio, so cost follows the length of the recording and not the difficulty of it. What separates them is languages covered, whether they mark who is speaking, and whether they run in real time or only on a finished file — a model that is excellent on a podcast may be unusable on a live call.
Weights published for research and explicitly not for selling. They are here because they exist, are often excellent, and are worth knowing about — and because the cheapest way to learn this is not after you have shipped. If a model on this list is what you want, the maker will usually license it commercially on request; that is a conversation, not a download.
- CrisperWhispernyra labs
- canary-1bNVIDIA
- diar_sortformer_4spk-v1NVIDIA
- mms-1b-allfacebook
- seamless-m4t-largeAI at Meta
- seamless-m4t-v2-largeAI at Meta
- wav2vec2-large-robust-12-ft-emotion-msp-dimaudeering
- xeusESPnet