Dlulisa bese ususa ihele lakho lapha
noma bheka amafayela akho
Imahhala: {0} amafayela ngosuku · kuze kube yi {1} min & {2} MB ngayinye
I-drafting ephilayo
Amathiphu avela njengoba ukhuluma, abhalwe phezulu yi-AI yethu — chofoza ubhale phezulu uma uqedile ukufaka inguqulo ehlanzekile, ehleliwe.
Ukungena kwe-microphone kulahliwe - vumela lokhu kuzinhlelo zendawo yesiphequluli sakho, noma ulayishe ifayela endaweni.
Ohele abakhulu noma uku-upload okuningi?
I-akhawunti emahhala: amafayela angama-5/izinsuku, amafayela angamahora angama-1
·
Pro: amafayela angamahora angama-10 + ama-labels omsindo
Ukulayisha...
Gcina lesi sithonjana sivulekile — uzodluliselwa ku-transcript yakho.
Talks & speechesReal-time notesIn-room speechAccessibility
Ngokungafani: ukukhuluma kwenza ukukhuluma nge-AI yethu ngesikhathi ukhuluma
Uma urekhoda, ikhasi lidlulisa umsindo we-microphone ibe ngama-segments ancane — amasekondi ambalwa ngalinye, licutshungulwa ngokunembile ukuze i-microphone elandelayo ihambe ngesikhathi esifanayo lapho i-microphone edlule ibuyela khona — futhi lidlulisa wonke ama-segments aqondile kumaseva ethu we-GPU, lapho imodeli efanayo ye-Whisper-class enamandla okufaka i-site yonke ibhala khona ngokushesha. Ubhalo luya emingwane ephilayo elandela umsindo wakho futhi luhlala luyaqhubeka kuze kube sengathi ukhuluma. Akukho inkonzo yokukhuluma ye-browser, akukho ingxenye yesithathu yokuchaza API — umbhalo ophilayo u-Whisper ovela kuma-hardware ethu.
Ngenxa yokuthi idinga kuphela ukurekhodwa kwe-microphone, indlela yokuphila isebenza kuwo wonke umsakazo omusha — i-Chrome, i-Edge, i-Safari, ne-Firefox, idesktop nefoni. Isiqephu ngasinye somsindo sisuswa kumaseva ethu ngesikhathi umbhalo wayo ubuyela emuva.
Ukudlalwa okuqhubekayo vs. ukudluliswa kokuphela
Ukudluliswa kwengxenye-nge-ingxenye kunemikhawulo efakwe ngaphakathi: imodeli ibona imizuzwana embalwa ngasikhathi sinye, ngakho-ke igama elidlula izingxenye ezimbili lingaphuma licutshungulwe, futhi ukuphawula phakathi kwezilimi zengxenye kuyinkokhelo. Lokhu kusho ukuthi ukuphuma kuphele ngenye indlela: ukurekhodwa okuphelele kugcinwa kwi-browser yakho uma ukhuluma, futhi ukucindezela ukushicilela kuqhuba ngemodeli njengento eyodwa — isihloko esigcwele, ukuphawula okulungile, ama-timestamps, umhleli, kanye ne-TXT/SRT/VTT ephumayo.
Sebenzisa umngenisi ophilayo ukuqinisekiswa ukuthi ihlala ikhona futhi uthathe umbhalo ngesikhathi esikhuluma (kunazinkinobho zokukopisha kumngenisi); sebenzisa isingeniso esiphelile nganoma yini ongayigcina. Uma ukhuluma umsindo okhona njengefayela, thumela indlela kunoma iyiphi ikhasi lethuluzi yindlela encane.
Imibuzo ebuzwa kaningi
Ingabe lokhu kubukeka kumnandi, noma kurekhodiwe futhi kuthunyelwe ngemuva?
Ngokungafani nokuphila: ukurekhoda nokudlulisa okudlulisa. Ikhasi lithumela wonke amasekondi ambalwa ambalwa we-audio ku-GPUs ethu ngenkathi ukhuluma olandelayo, ngakho umbhalo uqoqa ngesikhathi sokurekhoda, hhayi ngemuva. Lindela ukuthi u feed uphume ngesekondi ezincane ngemuva komsindo wakho — isikhathi esithatha isiqephu ukuqedela, ukufaka, nokudlulisa.
Kungani umbhalo ophilayo uhlala esigabeni esisodwa ngemuva kwami?
Izinqumo ezintathu ezincane zihlanganiswa: isiqephu kufanele siqedwe ukurekhodwa (imizuzu emithathu noma njalo), sihambe kuseva, futhi sidlulele ngemodeli. Okuvame ukubeka isiqukathi cishe imizuzu engu-4-8 ngemuva kwezwi lakho — futhi uma i-GPU yethu isebenza kancane, ikhasi lilungele ukuthunyelwa kwesixhumanisi eside kakhulu endaweni yokushiya emuva. Ukukhuluma okujwayelekile ngezikhathi zokuphumula kubukeka kuhlanzekile.
Yiziphi iziphequluli ezisebenza ngendlela ephilayo?
Zonke ezimanje — Chrome, Edge, Safari, ne Firefox, kudesktop nakumafoni. Kusekelwe kurekhodi le-microphone elijwayelekile ngaphezu kwesevisi yokukhuluma ye-browser, ngakho isebenza ku-Firefox njenge-Chrome.
Ingabe umsindo wami uhamba lapho ukuguqulela okuzenzakalelayo kusebenza?
Ku-servers zethu kuphela, futhi ngokucacile kucacisiwe: indlela yokuphila idlulisa umsindo wakho kuthi kungxenyekazi ezincane njengoba ukhuluma — lokho kwenza kube mnandi — futhi ingxenye ngayinye isuswa ngokushesha ngemuva kokuthunyelwe kwayo. Akukho lutho oluya ku-Google, Apple, noma iyiphi i-akhawunti ye-3rd party. Urekhodi oluphelele luhlala kwi-browser yakho kuze kube ucindezela Ukudlulisa, futhi ukudluliswa okungaziwa kususwa ngokuzenzakalela ngaphakathi kwehora le-24.
Ingabe ngingayisebenzisa ukulandelela inhlanganiso noma isifundo esikhona egumbini?
Njengesizo lomuntu siqu, yebo — kubeka isisetshenziswa eduze komsindo futhi ukudla kulandelela ukukhuluma ngokucacile kahle, cishe amasekondi ayishumi emuva. Yazi izimo: i-microphone eyodwa, ngakho-ke imisindo ephakathi nemisindo ikhuphuka, futhi akuyona izinsizakalo eziqinisekisiwe zokudlulisa izihloko zokufinyeleleka. Chofoza Ukuguqulela ekugcineni futhi i-full-context pass ivame ukubuyisela lokho okudliwayo okuphilayo.
Yini ulwimi olusebenza ngokuzijabulisa, futhi kufanele ngikhethe elinye?
Iningi lezinhlamvu ezingu-90+ njengezinhlamvu zonke ezibhalwe kulo mbhalo — yisimo esifanayo. Isibonelo esikhethekile sohlelo lokuphilayo: khetha ulwimi lwakho ku-selektor endaweni yokushiya ukuhlola okuzenzakalelayo. Ukuhlola kuqhutshwa ngalesi siqephu, futhi imizuzwana eyisikhombisa ayikho bufakazi obuningi, ngakho-ke ukubopha ulwimi kugcina u feed usuka ekungalinganini kwezilimi ezincane noma eziseduze.