I-Gemini 3.5 Transcribe: Imodeli entsha ye-Google yokushintsha izwi libe umbhalo

  • I-Google yembula i-Gemini 3.5 Transcribe, imodeli yayo yokukhuluma ibe umbhalo enembile kakhulu kuze kube manje.
  • Inikeza umbhalo ohlakaniphile olungisa ukulungisa okuzenzakalelayo, ususe amagama okugcwalisa, futhi ufometha umbhalo ngokuzenzakalelayo.
  • Isekela izilimi ezingaphezu kuka-85, ikwazi ukuqaphela izikhulumi ezintathu, futhi inciphisa isikhathi sokubhala ngo-70% uma kuqhathaniswa ne-Chirp 3.
  • Isivele itholakala ku-Gboard, Gemini ye-macOS kanye nonjiniyela; ukufika kwayo ku-Chrome kuhleliwe.

Ukubhalwa kwe-Gemini 3.5

I-Google iveze i-Gemini 3.5 Transcribe, imodeli yayo entsha ye-AI egxile ekuguquleni inkulumo ibe umbhalo. Le nkampani iyichaza njengohlelo lwayo olunembe kakhulu kuze kube manje, olwenzelwe ukusebenzisana ngezwi okuhlakaniphile. Ngokungafani nezinhlelo zendabuko, le modeli ayigcini nje ngokubhala lokho okushiwoyo, kodwa futhi ichaza inhloso yesikhulumi, ilungise amaphutha ngokushesha, futhi ikhiqize umbhalo ohlanzekile, ofomethiwe.

Imodeli yenzelwe ukusetshenziswa ngesikhathi sangempela kanye nokuqoshwa, futhi ithembisa ukusebenza kahle ezindaweni ezinomsindo, ngezimpawu ezinamandla, noma nge-jargon yobuchwepheshe. Ngaphezu kwalokho, ithola ngokuzenzakalelayo izilimi ezingaphezu kuka-85, ibone izikhulumi ezifika kwezintathu, futhi ingeze izitembu zesikhathi zegama negama. Konke lokhu kufezwa ngesilinganiso samaphutha esiphansi kakhulu, ngokusho kokuhlolwa okuzimele.

Ukubhalwa okuhlakaniphile: okungaphezu nje kokubizela

Ukubhalwa kwe-Gemini 3.5

Esinye sezici ezibalulekile ze-Gemini 3.5 Transcribe yikhono layo lokuqonda ukuzilungisa. Isibonelo, uma isikhulumi sithi, "Masihlangane ngoLwesibili... cha, ngoLwesithathu," imodeli izosebenzisa inketho yesibili. Iphinde isuse amagama okugcwalisa afana no-"uh" noma "um" futhi isebenzise ngokuzenzakalelayo izimpawu zokubhala nokufometha. Ngaphezu kwalokho, ikuvumela ukuthi wengeze isilulumagama esenziwe ngokwezifiso ukuze uqaphele amabizo afanele, amagama obuchwepheshe, noma amagama angavamile. Ngokusho kwedatha eshicilelwe yi-Google, izinga lamaphutha ngegama ngalinye lingu-4% wokusakaza umsindo kanye no-2,6% womsindo oqoshiwe ngaphambilini, njengoba kulinganiswa yi-Artificial Analysis.

Esinye isici esivelele isivinini. Uma kuqhathaniswa ne-Chirp 3 eyandulelayo, i-Google ithi isikhathi sokuthola umbhalo wokugcina sincishiswe ngo-70%. Ekusakazeni bukhoma, ukubambezeleka kungaphansi komzuzwana owodwa, okuvumela ukusetshenziswa kwayo ekubhaleni amagama ngesikhathi sangempela noma kubasizi bezwi.

Ukuhlanganiswa ohlelweni lwe-Google

Ukubhalwa kwe-Gemini 3.5

I-Gemini 3.5 Transcribe isivele ihlanganiswe nemikhiqizo eminingana ye-Google. Ku-Android, isici se-Gboard se-Rambler sisebenzisa le modeli yokufaka i-voice-to-text. Siyatholakala futhi kuhlelo lokusebenza lwe-Gemini lwe-macOS, okuvumela abasebenzisi ukusebenzisa imiyalo yezwi enomongo wesikrini ukufingqa amafayela, ukukhiqiza izithombe, noma ukusebenzisa kabusha umbhalo. Ngaphezu kwalokho, i-Google Antigravity iyifaka ukuhlanganisa umongo wesikrini nomlando wengxoxo.

Le nkampani iphinde yamemezela ukuthi lesi sici sizotholakala maduze ku-Chrome. Lokhu kuzovumela abasebenzisi ukuthi babize umbhalo kunoma iyiphi insimu yewebhu, njenge-imeyili, amafomu, noma okuthunyelwe, ngaphandle kokuthayipha. Nakuba kungekho usuku oluqondile, lesi sici kulindeleke ukuthi sandise ukusetshenziswa kokubhalwa kwezwi kusiphequluli.

Ukutholakala konjiniyela

Ukubhalwa kwe-Gemini 3.5

Kwabathuthukisi, i-Gemini 3.5 Transcribe iyatholakala ekubukweni komphakathi nge-Gemini API, ku-Google AI Studio, kanye naku-Gemini Enterprise Agent Platform. Kunezinhlobo ezimbili ezinikezwayo: i-Live API yokusakaza bukhoma ngesikhathi sangempela, okuqondiswa kabili nge-latency engaphansi kwesekhondi elilodwa, kanye ne-Interactions API yokucubungula umsindo oqoshiwe kusengaphambili nge-speaker attribution kanye ne-timestamps. Eyokuqala yenzelwe ama-ejenti ezwi kanye nokubhala amagama abukhoma, kanti eyesibili iwusizo emihlanganweni, ezingxoxweni, noma ekuhlaziyeni izingcingo.

Imodeli ingaphinde inikeze imisebenzi eyinkimbinkimbi kwamanye amamodeli e-Gemini, njengokukhiqiza izithombe noma ukuhlaziywa kwamafayela, ngokusebenzisa izingcingo zomsebenzi. Leli khono liyatholakala okwamanje kuhlelo lokusebenza lwe-Gemini lwe-macOS.

Ukuzibophezela kwe-Google ekukhulumeni njenge-interface eyinhloko kuqiniswa yi-Gemini 3.5 Transcribe. Inhlanganisela yayo yokunemba, isivinini, kanye nezici ezihlakaniphile iyenza ithuluzi elibalulekile kokubili abasebenzisi kanye nabathuthukisi. Njengoba isifikile ku-Chrome, ukubiza ngezwi kungaba yinto evamile kuwebhu. Kusazobonakala ukuthi izozivumelanisa kanjani nezimo zokusetshenziswa ezidinga kakhulu, kodwa idatha yokuqala iyathembisa.


Engeza njengomthombo okhethwayo ku-Google