Distil-Whisper large-v3
Free.ai (self-hosted)
·
stt
·
~500 tokens per minute
Distil-Whisper large-v3 se a Modèl pale-a-tèks ki te bati pa HuggingFace. Li se pi gwo nan Real-time transcription, large-volume batch STT.. Self-hosted sou Free.ai GPUs — kouri gratis kont chak jou ou a token pool (500 tokens pou chak minit). Li libere anba MIT - itilize komèsyal ki pèmèt sou Free.ai.
Itilize via API
API REST ki konpatib ak OpenAI. Pwodui yon kle epi rele modèl sa a nan kèk segond.
curl -X POST https://api.free.ai/v1/stt/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"distil-whisper-large-v3","audio_url":"https://..."}'Kesyon ki poze souvan
Distil-Whisper large-v3 transkri son ki pale nan tèks. Upload yon MP3, WAV, M4A, oswa videyo dosye ak Distil-Whisper large-v3 retounen transkript konplè plus optional SRT / VTT sous-titres ak timestamps.
Distil-Whisper large-v3 handles dozens of languages - Whisper-family models cover 90+, others vary. Pick "auto-detect" or specify the language for highest accuracy.
Rapò erè mo se 5-10% sou son an Angle pwòp, 10-20% sou son an briyan oswa accenté. gwo varyasyon nan menm achitekti fè pi byen sou ka difisil - chwazi pi gwo lè son an se amann.
Wi - chak segment gen ladann kòmanse/fini timestamps. Ekspòtasyon kòm SRT oswa VTT ak tan mapin dwat sou ou videyo.
Distil-Whisper large-v3 kouri sou GPUs nou kont dlo gratis chak jou ou an premye; $5 → 200,000 tokens peye apre sa. Avèk ~ 500 tokens pa minit.
MP3, WAV, M4A, FLAC, OGG, plus videyo (MP4, MOV, WebM) - nou ekstrè son an. Max 500 MB pou chak upload. Pi long dosye? Divizyon an ak /audio/cut/ oswa itilize /v1/stt/batch/.
Speaker diarization se yon pase separe - toggle "diarize" sou / transcribe /. Distil-Whisper large-v3 jere transkriptyon an; diarization etikèt chak segment ak Speaker 1 / Speaker 2 / etc.
Wi - /batch/ aksepte yon dosye ki gen fichiers son. Chak transkripte fini nan /account/?tab=history ak non dosye orijinal la. Pou konsève arbr-dosye a, itilize API a.
Wi - POST ou son an nan /v1/stt/transcribe/ ak modèl "Distil-Whisper large-v3". Retounen JSON ak tèks + segments + word-level timestamps. /api/ gen referans konplè.
Modèl ki fè tèt yo rete kenbe son sou GPU nou yo; premium pase pa yon DPA. Son an se efase apre fenèt pataje (24h anon, 7d signed-in). Nou pa antrene sou ou enpòte.
Wi - Free.ai bay dwa pou itilize transkripsyon an pou rezon komèsyal. Ou bezwen dwa pou son an ou mete ajou (youn nan enregistrement ou, materyèl ki gen lisans, oswa kontni ak konsantman).
Faktori tan reyèl la se sou 0.05-0.2× - yon podcast 60-minute transkripte nan 3-12 minit. Premium modèl yo souvan fini pi vit.