
Ibhulogi yezobuchwepheshe i-TestingCatalog ishicilele umbiko ngo-Agasti 2, ithi iMicrosoft ihlola imodeli yayo yokuqala yezwi yesikhathi sangempela, i-MAI Realtime, esekela izilimi ezingu-16 namazwi angu-2, futhi ekwazi ukulalela nokukhuluma ngesikhathi esisodwa engxoxweni, ibone ulimi ngokuzenzakalelayo futhi ishintshe ulimi phakathi nengxoxo.
Mayelana nokuhlolwa, umbiko uthi iMicrosoft isimeme abanye ozakwethu ukuthi bahlole le modeli endaweni i-MAI Playground. Okwamanje akukacaci ukuthi izotholakala nini ku-Microsoft Foundry, noma nini ezonwetshwa iye emsebenzini wezwi we-Copilot.
Mayelana nokusebenza, i-MAI Realtime isebenzisa ukuxhumana okubheke ezinhlangothini zombili, okusebenza ngesikhathi esisodwa. Akudingeki ukushintshana okuqinile kokuthi “umsebenzisi aqede ukukhuluma, bese imodeli iphendula”; ingalalela futhi iphendule ngesikhathi esisodwa.
Mayelana nezilimi ezisekelwayo, imodeli ye-MAI Realtime isekela izilimi ezingu-16, okuhlanganisa isiShayina. I-IT之家 icaphuna ibhulogi futhi inikeze uhlu lwezilimi ezisekelwayo kanje:
IsiNgisi
IsiJalimane
ISpanishi
IsiFulentshi
IsiNtaliyane
IsiPutukezi
IsiJapane
IsiKorea
IsiShayina
IsiDashi
IsiHindi
Isi-Indonesian
Isi-Arabhu
IsiRashiya
IsiTurkish
IsiVietnamese
IsiThai
Mayelana nezitayela zezwi, umbiko uthi ukuhlolwa kwale modeli kunikeza amazwi amabili, i-Victoria ne-Grant, futhi azwakala ngokwemvelo kakhulu kunezindlela zezwi ezitholakala manje ku-Copilot. Abasebenzisi bangacacisa ulimi ngokwabo noma bavule ukutholwa okuzenzakalelayo.


Le modeli ayisekeli ukucula noma ukukhiqiza imisindo engewona owezwi, futhi isagxile ekubeni uhlelo lwezingxoxo. Iphaneli yokulungisa ingabonisa ukubambezeleka kwesikhathi sangempela, okuqukethwe imodeli ecabanga ngakho nezinyathelo zokucubungula. Umsebenzi wokwabelana ngamasampula ungase uvulelwe abasebenzisi benkundla lapho imvume yokuhlola isiyandisiwe.

Qaphela: Izwi elisebenza ngezindlela zombili ngesikhathi esisodwa (Full-duplex voice) lisho ukuthi uhlelo lungamukela izwi lomsebenzisi futhi lukhiphe impendulo ngesikhathi esifanayo, ngaphandle kokulinda ukuthi olunye uhlangothi luqede ngokuphelele ukukhuluma. L依thembela ekutholweni kwamaphuzu okuphela kwenkulumo ukuze libone ukuqala kokukhuluma, ukuphumula nokuphela kwalo, futhi kufanele lilungise okukhiphayo lapho umsebenzisi engenela phakathi.
Amamodeli ezwi e-MAI akhishwe yiMicrosoft ngaphambilini wonke ayesebenza ngendlela eyodwa, okuhlanganisa i-MAI-Voice-2 kanye nenguqulo yayo ye-Flash esetshenziselwa ukuhlanganisa izwi, kanye ne-MAI-Transcribe-1.5 esetshenziselwa ukubona izwi.
