Kwenzekeni
I-Google DeepMind imemezele ukukhishwa kwezinhlobo ezimbili ezintsha zokuguqula umbhalo ube inkulumo, i-Gemini 3.8 Flash TTS kanye ne-Gemini 3.8 Flash-Lite TTS. Lawa mamodeli atholakala ngokushesha ku-Google AI Studio nange-Gemini API, ngokuhlanganiswa kwamapulatifomu afana ne-Agora, i-LiveKit, ne-Vercel. Ukukhishwa kwandisa umndeni womsindo we-Gemini, kwengeza amakhono okukhiqiza amazwi abalingiswa bangokwezifiso nokuqondisa inkhulumomphendvulwano yesigcawu ngokulawula okunembayo kokuletha.
I-Google DeepMind yethule i-Gemini 3.8 Flash TTS kanye ne-Gemini 3.8 Flash-Lite TTS, iwachaza njengamamodeli esizukulwane somsindo azwakalayo kakhulu emndenini we-Gemini. Isimemezelo sithi lawa mamodeli aguqula ukukhiqizwa kwezwi kusuka ekusetheni kusengaphambili okumile kuya kusitudiyo sokudala esiguqukayo, okuvumela ukudalwa kwamazwi abalingiswa bangokwezifiso kanye nesiqondiso sengxoxo yesigcawu.
Amamodeli ayatholakala kusukela namuhla ku-Google AI Studio, lapho esebenza njengendawo yokusebenza yokuklama izwi. Abasebenzisi bangakwazi ukwazisa ubunikazi bephimbo obusha kusukela ekuqaleni noma baphindaphinde amazwi abo, bese besebenzisa umhleli wesikrini onezipikha ezimbili ukuqondisa ukulethwa komugqa ngomugqa. Ukufinyelela futhi kunikezwa nge-Gemini API, evumela izinkundla zonjiniyela ezifana ne-Agora, i-LiveKit, i-Pipecat, ne-Vercel ukuze zakhe futhi zikhiphe ukuzizwisa kokukhiqiza inkulumo.
I-Google ithi i-Gemini 3.8 Flash TTS ivikela indawo engu-#1 iyonke ku-Hume AI's Voice Design ngamaphuzu angu-71.4 futhi izihola phambili ngemodeli ye-accent ngamaphuzu angu-60.8. Womabili amamodeli kubikwa ukuthi avikele indawo engu-#1 kanye ne-#2 ku-Hume AI's Overall Quality Index. Ekuhlaziyeni okuthandwayo komuntu okungaboni ku-Voice Arena, amamodeli kuthiwa abambe izikhundla eziphezulu ezilimini ezibalulekile zomhlaba wonke ezihlanganisa isiJapanese, isiPutukezi saseBrazil, isi-Vietnamese, i-Modern Standard Arabic, i-Mexican Spanish, nesi-Hindi, ngokusekelwa kwezilimi ezingaphezu kwe-100.
Ukukhishwa kuhlanganisa izindlela ezithile zokuphepha zokuphindaphinda kwezwi, okudinga abasebenzisi ukuthi banikeze imvume yomlomo eqoshiwe evela kumnikazi wezwi efana nesipikha esiyisithenjwa ngaphambi kokuthi kudalwe izwi. Ukwengeza, zonke iziqeshana zomsindo ezikhiqizwa yilawa mamodeli zimakwa nge-SynthID, i-watermark engabonakali eklanyelwe ukuqinisekisa ukuthi inkulumo ekhiqizwe yi-AI ihlala ibonakala ukuze kusize ukuvimbela ulwazi olunganembile.
Imininingwane yomthombo: deepmind.google ↗
Kungani kubalulekile
Lokhu kwethulwa kuphawula ushintsho olukhulu esizukulwaneni sezwi se-AI sisuka kokusethiwe okumile siye ekudalweni komsindo okuguquguqukayo, okwenziwa ngokwezifiso. Ngokunika amandla onjiniyela ukuthi bakhe izinkomba zephimbo ezintsha ngokuphelele noma baphindaphinde amazwi athile ngokuqinisekisa kwemvume, amamodeli avula amathuba amasha okwenziwa kwasendaweni kwemidiya, ama-ejenti ezingxoxo, nokukhiqizwa kokuqukethwe okunobuhlakani. Ukufakwa kwe- ye-SynthID kubhekana nokukhathazeka okukhulayo mayelana nolwazi olungaqondile olukhiqizwa yi-AI, okuhlinzeka ngesivikelo sobuchwepheshe sokubonisa ngale kokuqukethwe.
Ukwethulwa kwalawa mamodeli kunikeza onjiniyela namabhizinisi ngamathuluzi okudala umuzwa ocebile, ozwakalayo womsindo ngaphandle kokuthembela ekusetheni kwezwi okungashintshi. Lawa mandla afaneleka kakhulu ezimbonini ezidinga ukuphimisela kwesifunda okuhlukile ukuze kwenziwe kwasendaweni imidiya noma amazwi ebhrendi angaguquki kubasebenzeli bezingxoxo.
Ubambiswano nezinkampani ezifana ne-Figma, i-HeyGen, i-Linguana, i-Wondercraft, i-99.co, ne-Ollang ikhombisa ukusebenza ngokushesha okusebenzayo ekusheshiseni ukuqoshwa komhlaba wonke kanye nokunika amandla ama-ejenti ezwi ezingxoxo ngezinga elikhulu. Lokhu kuphakamisa umnyakazo obheke ekuhlanganiseni amakhono e-TTS athuthukile kuzinhlelo zokusebenza ezijwayelekile zokudala kanye nebhizinisi.
Ukugcizelelwa kwemvume yokuqinisekiswa kokuphindaphindwa kwezwi kanye nokusetshenziswa kwe- ye-SynthID ibhekana nokukhathazeka okubalulekile kwezimiso zokuziphatha nokuvikeleka ekwenziweni komsindo we-AI. Lezi zivikelo zihlose ukuvikela ubunikazi bethalente lezwi futhi kuqinisekiswe ukubonakala kokuqukethwe, okubaluleke kakhulu njengoba imidiya ekhiqizwa yi-AI ivame kakhulu.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').In AI, what are a model's "parameters"?
Ongakubuka ngokulandelayo
Gada ukwamukelwa kwalawa mamodeli yizinkampani zozakwethu ezifana ne-Figma ne-HeyGen ukuze kubhalwe phansi umhlaba wonke kanye nokwenza imidiya ibe yasendaweni. Buka ukuhlolwa okuzimele kokuvikela ukuphindaphinda kwezwi nokusebenza kwe-SynthID ekutholeni inkulumo ekhiqizwe yi-AI. Ukwengeza, bheka ukuthi umhleli wesikrini onezipikha ezimbili ku-Google AI Studio usetshenziswa kanjani onjiniyela ekulandiseni okuyinkimbinkimbi komsindo.
Qaphela ukuthi izinkampani zozakwethu ziwahlanganisa kanjani lawa mamodeli emikhiqizweni yazo, ikakhulukazi ezindaweni zokuqopha umhlaba wonke kanye nokwenza kwasendaweni imidiya, ukuze kuhlolwe ukusebenza komhlaba wangempela nokwamukela umsebenzisi.
Gada ukuhlolwa okuzimele okuvela eceleni kwezindlela zemvume yokuphindaphinda izwi kanye nokutholakala kwama-watermark e-SynthID, njengoba lezi zibalulekile ekuqinisekiseni ukuphepha nobuqotho bobuchwepheshe.
Landela ngomkhondo ukuthuthukiswa komhleli wesikrini esinezipikha ezimbili ku-Google AI Studio, njengoba lesi sici simelela isixhumi esibonakalayo esisha sokuqondisa ingxoxo ekhiqizwe yi-AI engathonya indlela abadali ababhekana ngayo nokuxoxwa kwezindaba ezilalelwayo.