Distil-Whisper large-v3
Free.ai (self-hosted)
·
stt
·
~500 Token kwa minute
Distil-Whisper large-v3 bụ a ngwe-ka-okwu móòdù e mepụtara site na HuggingFace. Ọrụ na Real-time transcription, large-volume batch STT.. Self-hosted na Free.ai GPUs - na-agba ọsọ n'efu megide ụbọchị gị token pool (500 tokens nkeji). E wepụtara ya n'okpuru MIT — iji azụmahịa ekwenyela na Free.ai.
Jiri site na API
OpenAI-na-akpaghị aka REST API. Kewapụta kii nakwa kpọọ móòdù a n'ime sekọnd.
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://..."}'Ajụjụ ndị a jụrụkarị
Distil-Whisper large-v3 na-atụgharị ụda akọwapụtara n'ime ngwe. Bipụta faịlụ MP3, WAV, M4A, mọọbụ faịlụ vidio na Distil-Whisper large-v3 na-eziga ntụgharị zuru ezu nakwa SRT/VTT n'aka ekpe na oge.
Distil-Whisper large-v3 na-ejikwa asụsụ ndị dị iche iche - Whisper-family models cover 90+, Parakeet covers ~25, ndị ọzọ na-agbanwe. Họrọ "auto-detect" mọọbụ kọwaa asụsụ maka nghọta dị elu.
Ọnụọgụgụ okwu-ezighị ezi bụ 5-10% na ụda English dị ọcha, 10-20% na ụda na-anụ ọkụ ma ọ bụ ụda na-akụda. Ụdị dị ukwuu nke ụda ahụ na-eme ka ọ dị mma n'ihe ndị dị njọ - họrọ nnukwu mgbe ụda ahụ dị njọ.
Ee - segmenti ọbụla na-agụnye oge mbido/nkewapụta. Ekpughe ya dịka SRT mọọbụ VTT na oge map n'ụzọ ziri ezi na vidyo gị.
Distil-Whisper large-v3 na-agba ọsọ na GPUs anyị onwe anyị megide pool gị n'ụbọchị mbụ; $5 → 200,000 tokens na-akwụ ụgwọ mgbe ahụ. N'ihe gbasara ~ 500 tokens kwa nkeji.
MP3, WAV, M4A, FLAC, OGG, nakwa vidiyo (MP4, MOV, WebM) - anyị na-ewepụ ụda. Max 500 MB kwa ibudata. Files dị ogologo? Split na /audio/cut/ ma ọ bụ jiri /v1/stt/batch/.
Nhazi nke onyeọsụsụ bụ nbanye dị iche iche - gbanwee "diarize" na /transcribe/. Distil-Whisper large-v3 na-ejikwa ntụgharị; nhazi nke onyeọsụsụ na-egosipụta segmenti ọbụla na Onyeọsụsụ 1 / Onyeọsụsụ 2 / wdg.
Ee — /batch/ na-anabata nsomebefaịlụ nke faịlụ ụda. Nhazi nke ọbụla na-abanye na /account/?tab=history na aha faịlụ mbụ. Maka nchekwa nke nsomebefaịlụ-tree jiri API.
Yabụ - POST ụda gị na /v1/stt/transcribe/ na model="Distil-Whisper large-v3". Na-eziga JSON na ngwe + segments + oge-ihe dị na okwu. /api/ nwere ndesịta zuru ezu.
Models na-echekwa onwe ha na-echekwa ụda na GPUs anyị; ọbụna na-aga site na DPA. Ọdịdị a na-ehichapụ mgbe windo n'otu (24h anon, 7d banye-na). Anyị anaghị arụ ọrụ na init gị.
Ee — Free.ai na-enye ikike iji n'ọrụ n'ọrụ nke transcripts. Ichọrọ ikike n'ihe oyiyi ịgbapụtala (n'ihe oyiyi gị, ihenhọrọ nke ikike, mọọbụ ihenhọrọ nke ikike).
Real-time factor bụ ihe dị ka 0.05–0.2× — 60-minuite podcast transcribes na 3–12 minit. Premium models mgbe ụfọdụ na-agwụ n'ụzọ n'ụzọ. Jiri bọtịnụ n'okporo ụzọ mechie táàbụ̀ ahụ.