Distil-Whisper
Kāore e whakahaeretia e te STT.ai tēnei tauira. Ko tēnei pātaka he mōhiohio tohutoro mō ia.
Ka whakamātautau ki ngā tauira e whakahaeretia ana e tātau →
5.8%
WER
99
reo
48.0x
Āhuatanga
MIT
Whakawhiwhinga
Mo Distil-Whisper
Ko te Whisper-Whisper he putanga whakawātea o te Whisper i hangaia e Hugging Face. Ka whakaiti i te rahi tauira e 49% ā, ka whiwhi whakamāramatanga tere ake e 6x i te wā e pupuri ana i roto i te 1% WER o te Whisper Large V2 taketake i ngā huinga arotakenga i waho.
Whakamāramatanga o te tauira
- KaiwhiwhiHugging Face
- Hanganga-
- WhakawhiwhingaMIT
- WhakahōuMar 2026
E pā ana ngā pātai
Ko te Distil-Whisper he tauira kōrero-ki-tuhi e te Hugging Face. Kāore e haere te STT.ai i tēnei wā Distil-Whisper — ko tēnei pātaka he mōhiohio tohutoro mō te tauira. Mō te whakamārō i te STT.ai, ka tuku te tīpako kōwhiria i te whānau Whisper (Turbo, Large V3, Medium, Small).
I runga i ngā tohutoro paerewa, e tae ana te Distil-Whisper ki te 5.8% Wāhi hapa Wā. E whakawhirinaki ana te tika o te ao tūturu ki te āhuatanga oro, ki te āhua, me te reo; mō ngā pūkete mārō, whakahua rānei, e tūmanako ana ki ētahi wāhanga ōrautanga tiketike ake i te WER.
Kāore e wātea ana te Distil-Whisper ki te STT.ai. Ko ōna tikanga whakaaetanga ake e whakahaere ana i te whakahaere i a koe. Ko te taumata wātea o te STT.ai e taupoki ana i ngā tauira Whisper e whakahaere ana tātau — 600 ngā minu hei tīmata, kātahi ka wātea tētahi whakahau wātea i ia marama.
Ka tukua te Distil-Whisper i raro i te MIT, he whakaaetanga pūtake tūwhera. Ka taea e koe te whakawhiwhi i te Distil-Whisper ki o koe ake ngā pūrere, te whakamahi rānei i tātau putanga whakawhiwhia — he whai hua rānei.
E tautoko ana te Distil-Whisper i ngā reo 99. Ka kōwhiria e te kite-māori te reo tika mō te nuinga o te oro; Ka taea hoki e koe te whakapūtā i te ringa mō tētahi whakahau tika iti.
Distil-Whisper e mahi ana i te oro i te wā tūturu 48.0x i runga i a tātau GPUs. Ka oti te pūkete oro 1-ora i raro i te 1 minu; he roa ake te raupapa o ngā pūkete, ā, ka whakamōhiotia mā te imeli ina oti.
He tika ake ngā tauira nui ake, engari he pōturi ake; Ko te STT.ai e noho ana i te Distil-Whisper i runga i te GPU kia kore ai te tatau parameter e pā ki tōna mahi taha o te kaiuru.
E whakaae ana te Distil-Whisper ki ia momo STT.ai e tautoko ana — MP3, WAV, M4A, FLAC, OGG, MP4, MKV, MOV, WebM, AVI, me ētahi atu. Ko te huaputa he TXT, SRT, VTT, DOCX, JSON, PDF rānei.
He. Ka haere te whakahua kōrero ki te Distil-Whisper mō ia whakahua - ka whakahuatia ia kaikōrero, ā, ka taea e koe te whakahōu i a rātou i te kaiwhakatika i muri iho.
He. Distil-Whisper e haere ana i roto i tātau taiao whakahaere - ka tukatuka, ka whakakoretia te oro, ā, kāore anō kia whakamahia mō te whakaakoranga me te kore kōwhiringa mārama. Ka tāpiri ngā mahere Pro ki te whakapūtātanga o te taha o te kaiuru mō ngā tāruatanga i te wā e noho ana.
Ka whakamahia te utauta o te whakataurite-stt hei whakahaere i te Distil-Whisper ki ētahi atu tauira tautoko i runga i te oro ōrite — ka kitea e koe te WER, te tatau wāhanga, ngā tohu kaikōrero, me ngā pūkete ātete. Ko te whakataurite Distil-Whisper vs Whisper Large V3 te tino pūnoa.
He. Ka whakapūtātia "distil-whisper" hei tātai tauira i runga i te /v1/transcribe te wāhi mutunga. Ko ngā SDKs Python me Node.js kei roto ko ngā tauira Distil-Whisper. Kei roto i te taumata API wātea ko te 100 minu/whā.
He. Nā te mea he Distil-Whisper te MIT-licensed, ka taea e koe te whakanoho i a ia. Ko te STT.ai te pou pūtake tūwhera e whakarārangi ana i te pūnaha me ngā taumahatanga. Ko te nuinga o ngā rōpū whakanao e whakamahi ana i tātau putanga whakanoho hei whakarerekē i te GPU, i ngā whakarerekētanga tauira, me ngā taumahi.