Free IsiJapan Transcription
Bhala i-IsiJapan yesandi nevidiyo ibe ngumbhalo nge-AI. Ikhawulezayo, ichanekileyo, kwaye ikhululekile.
Indlela esebenza ngayo
- Yiya kwi- Free.ai Umshicileli
- Layisha phezulu ifayili yakho yesandi okanye yevidiyo IsiJapan
- I-AI yethu ifumanisa ngokuzenzekelayo IsiJapan kwaye ibhalela
- Layisha ezantsi i-transcript yakho njengombhalo okanye i-SRT subtitle
IsiJapan Iimpawu Zokushicilela
- ✓Isebenza nge faster- whisper (MIT licensed)
- ✓Ubhaqo oluzenzekelayo lweelwimi IsiJapan
- ✓Inkxaso ye MP3, WAV, MP4, M4A, FLAC, kunye nezinye
- ✓Ii-timestamps kunye norhwebo lwangaphandle lwesihloko esingaphantsi (SRT)
- ✓Akukho mda wobungakanani befayili kwinkqubo ehlawulweyo
- ✓Iifayile ezifihlakeleyo zigcinwa emva koqhubekeko
Iinkcukacha Zolwimi
| Iilwimi | IsiJapan |
| Ikhowudi ye-ISO | ja |
| Imodeli ye-AI | i-faster-whisper |
| Ixabiso | Iinketho zelizwe |
Iilwimi ezininzi
Bonisa Zonke IilwimiImibuzo ebuzwa rhoqo
I-Whisper enkulu-v3-i-turbo iwela kwinqanaba eliphezulu lempumelelo kwi IsiJapan - ngaphantsi kwe 7% yexabiso lemposiso yegama kwiimpawu eziqhelekileyo ze-benchmark. Kwimisebenzi ethetha ukuba i-audio yestudio ecocekileyo ibuyela kwi-perfect, kwaye i-audio yencoko isetyenziswa ngococeko oluncinci.(Inqanaba A, under 7% word error rate kwiseti yexabiso elifanelekileyo - sipapasha inqanaba elithembekileyo le-WER kunokuba sibhale izibhengezo zorhwebo.)
Ewe — IsiJapan uguqulelo lutsala ukusuka kwi-token pool yakho yosuku olusimahla kuqala. Isandi sibiza malunga ne-50 tokens ngomzuzu, ngoko ke i-pool yosuku olungaziwayo igubungela iiyure ezimbalwa zesandi ngosuku. Ii-akhawunti ezibhalisiweyo zifumana i-pool enkulu kunye ne-10,000 signup tokens. Emva koko, $1 ithenga i-750,000 tokens (~ iiyure ezingama-250 zesandi).
IsiJapan iincwadi ezibhaliweyo zibuyiselwa kwiskripthi esisemthethweni (UTF-8). IsiJapan umbhalo awunazo izithuba phakathi kwamagama ngokusemthethweni; i-diarization timestamps idibanisa iziqhoboshi eziqhelekileyo kwimijikelezo yomthumeli.
I-MP3, i-WAV, i-M4A, i-FLAC, i-OGG, i-OPUS, ne-WEBM zivunyelwa ngokuthe ngqo. Kwividiyo (MP4, MOV, MKV) sikhupha umkhondo wesandi kwiseva-ecaleni phambi kokuba siyithumele kwi-Whisper — awunakutshintsha nantoni na ngokwakho. Inkqubo efanayo nokuba ithetha ntoni na ulwimi lomntu, kubandakanya IsiJapan.
Ukhuphelo olungaziwayo luya kufikelela kwi-500 MB kwifayili nganye. Ii-akhawunti ezibhalisiweyo ziye kwi-2 GB. Ukuphela kwexesha alikho umda onzima - iifayile ezide ziqhutywa ngokuzenzekelayo (iifestile zemizuzu engama-30 ezinamathele) kwaye zidityaniswe kwakhona kwi-transcript epheleleyo enee-timestamps eziqhubekayo. Iiyure ezininzi IsiJapan zokukhuphela (ipodcasts, izifundo ezipheleleyo, iintlanganiso) zisebenza kakuhle.
Ewe - ushicilelo lwediary lomculi luyasebenza ngokumiselweyo kwi IsiJapan yonke. Imveliso ihlukaniswe njengeMculi 1 / Umculi 2 / Umculi 3 ngee-timestamps, ngoko udliwanondlebe, unxibelelwano lwepaneli, kunye neentlanganiso zeqela elininzi zibuyela emva zilabelwe. Ushicilelo lwediary luyasebenza kwimodeli eyahlukileyo kwaye lusebenza ngokufanayo kuzo zonke iilwimi esixhasayo.
Ewe — Cola i-URL kwi /transcribe/youtube/ ye-YouTube okanye /transcribe/podcast/ ye-podcast feeds (i-Apple, Spotify, RSS). Sikhuphela ezantsi isandi, siyiqhube nge-Whisper nge-language=ja, kwaye sibuyisele i-transcript ngee-timestamps kunye neelabels zomthumeli. I-IsiJapan eqhelekileyo iqulethe: iipodcasts, izifundo, izincoko, kunye nemixholo yeYouTube ekwifomu ende kwi-IsiJapan yimithwalo yomsebenzi eqhelekileyo esiyibonayo.
I-Whisper ibiza malunga ne-50 tokens ngomzuzu wesandi, ngoko ke urekhodo lweyure enye li ~3,000 tokens. $1 ithenga i-750,000 tokens, esebenza ngokumalunga neyure ezi-250 zesandi ngedola. Abaninzi babasebenzisi abachithanga nto - i-pool yemihla ngemihla ekhululekileyo iquka ii-clip ezimfutshane, ii-voice notes, kunye neepodcasts ezi-one-off.
Ewe — zombini i-segment-level (imizuzu nganye ~10-30) kunye ne-word-level timestamps zifumaneka. I-word-level yi-default ye-VTT/SRT subtitle export ngoko ke izihloko zihamba ngaxeshanye umgca-nge-mgca. Kwi-API misela i-timestamps="word" kwisiqu sesicelo. IsiJapan iincwadi ezibhaliweyo zibuyiselwa kwiskripthi esisemthethweni (UTF-8). IsiJapan umbhalo awunazo izithuba phakathi kwamagama ngokusemthethweni; i-diarization timestamps idibanisa iziqhoboshi eziqhelekileyo kwimijikelezo yomthumeli.
Ewe. UTHENGA umsindo (inxalenye eninzi/ifomu-data, igama lendawo "ifayile") kwi /v1/transcribe/ nge-language=ja — okanye ushiye i-parameter ye-language ukuze i-Whisper ikwazi ukuvavanya ngokuzenzekelayo. Ibuyisela i-JSON ene-transcript, ii-segments, ii-timestamps, kunye nee-labels zomthumeli. Ubhekiso olupheleleyo kunye ne-SDK snippets kwi /api/.
Ewe - xa uguqulelo lugqityiwe, nqakraza Gcina okanye uncamathisele umbhalo kwi /translate/. IsiJapan idibanisa nezinye iilwimi zonke esizixhasayo (200+). Kwiimini zengxoxo uguqulelo luya ku /summarize/; xa kuthelekiswa, thumela ku /voice/tts/ ukuvelisa isandi kwiilwimi ezilindelweyo.
I-Whisper iqeqeshwe kwi-680K yeeyure zesandi esingenasandi sehlabathi, ngoko ke IsiJapan ukudluliswa kwesandi kunamandla kakhulu kwingxolo yasemva, iibhedi zemiculo, kunye nokulinganisa umgangatho wefowuni. Ukuchithwa okunzima okanye izithethi eziliqela eziliqela ziya kubangela ukuba umgangatho ubuhlungu.Ukuba i-transcript ibuyela ingekhoyo, thumela i-imeyili kwi contact@free.ai ngefayili — siya kubuyisela i-token kwaye sijonge ukuba i-engine eyahlukileyo iphatha isandi sakho kakuhle.