Ukuhlanganiswa kwenkulumo: Amarobhothi angagcina eveza imizwa

ISIKWELETU SESITHOMBE:
Isikweletu sezithombe
iStock

Ukuhlanganiswa kwenkulumo: Amarobhothi angagcina eveza imizwa

Ukuhlanganiswa kwenkulumo: Amarobhothi angagcina eveza imizwa

Umbhalo wesihlokwana
Ubuchwepheshe bokuhlanganisa inkulumo buvula amathuba amasha ama-bots asebenzisanayo.
    • About the Author:
    • Igama lomlobi
      I-Quantumrun Foresight
    • December 29, 2022

    Isifinyezo sokuqonda

    Nakuba inkulumo ekhiqizwe ngomshini isinesikhathi eside ikhona, kungenxa yentuthuko yokuqaphela inkulumo kuphela nokukhiqiza lapho isiqala ukuzwakala njengerobhothi elincane. Ezinye izinkampani zisebenzisa i-voice synthesis kanye nentuthuko ye-cloning ukuze zifake imizwa (okungukuthi, iphimbo) enkulumweni ekhiqizwe umshini. Imithelela yesikhathi eside yokuhlanganiswa kwenkulumo ingase ihlanganise ukuphinda kuvezwe amazwi osaziwayo kanye nokuqukethwe okuyimfihlo okukholisayo nakakhulu.

    Umongo wokuhlanganisa inkulumo

    Inkulumo yokwenziwa ikhiqizwa umthombo ongeyena owomuntu (isb, ikhompuyutha) kuyilapho idala kabusha umsindo wezwi lomuntu. Lobu buchwepheshe babukhona kusukela ngawo-1930s lapho unjiniyela we-acoustic waseMelika u-Homer Dudley akha i-vocoder yokuqala (i-voice synthesizer). Kancane kancane, kwaqala ukuvela amasistimu asebenzisa i-Gaussian Mixture Models (GMM) ukuthuthukisa ikhwalithi yokuhlanganisa inkulumo, nakuba kungesona isivinini. Kodwa-ke, intuthuko ekufundeni okujulile (i-DL, indlela yokufunda yomshini) kanye nobuhlakani bokwenziwa (AI) bucwengisise ubuchwepheshe ukuze bukhiqize izingxoxo ezikholekayo nezizwakalayo zemvelo. Ukuhlanganiswa kwenkulumo kusekelwa ngokuyinhloko amanethiwekhi amabili e-neural ajulile (DNN): umbhalo-kuya-enkulumweni (TTS) nokuguqulwa kwezwi (VC). 

    Umbhalo-ube-inkulumo uguqula umbhalo ube yizwi, kuyilapho i-VC ingaguqula izwi lomuntu ukuze lilingise elomunye. Lawa ma-DDN amabili avame ukusetshenziswa kubasizi ababonakalayo, futhi angakha amazwi nezingxoxo ezinamagama amaningi. Ukuhlanganiswa kwenkulumo kungakha abanakekeli bamarobhothi abagcizelela kakhulu nabasizi basekhaya bedijithali abahlakaniphile. 

    Kodwa-ke, ubuchwepheshe bezwi bokwenziwa bungasetshenziswa futhi ekuhlaselweni kwe-cyber. Le misebenzi yokukhwabanisa ikopisha izigxivizo zezwi zabantu (amasampula ezwi agcinwe ngedijithali ukuze asebenze njengokuhlonza kwabo kwebhayomethrikhi) ukuze kungene amasistimu namadivayisi. Ukwenziwa kwezwi kungakhohlisa ozakwabo ukuthi banikeze amagama abo ayimfihlo kanye nolunye ulwazi lwenkampani olubucayi. Amazwi antshontshiwe noma akhiqiziwe angasetshenziswa futhi ekuhlaselweni kobugebengu bokweba imininingwane ebucayi lapho abantu bakhohliswa ukuthi bathumele imali noma bayidlulisele kuma-akhawunti athile asebhange.

    Umthelela ophazamisayo

    Ngo-2021, abacwaningi benkampani yezokuxhumana i-Hitachi kanye ne-University of Tsukuba yase-Japan bakha imodeli ye-AI engakwazi ukulingisa inkulumo efana neyomuntu, okuhlanganisa nezimpawu zemizwa ezehlukene ezisekelwe kumsindo. Inkulumo ihloselwe ukuzwakala njengomnakekeli ochwepheshe. Amamodeli afana nalawa enzelwe ukuthi asetshenziswe kumarobhothi noma kumadivayisi angase anikeze ubungane, usekelo, nesiqondiso kubantu ngabanye abaludingayo. Ithimba lifundise imodeli yalo ye-AI ngokuqala ngokuyiphakela ngezibonelo zenkulumo ethinta inhliziyo.

    Ngemva kwalokho, isiboni semizwa siyaqeqeshelwa ukuhlonza umuzwa, futhi imodeli yokuhlanganisa inkulumo iyathuthukiswa ukuze idale inkulumo ethinta inhliziyo. Isiboni semizwa sisiza ukuqondisa isididiyeli senkulumo kuye ngokuthi imuphi umuzwa noma “imizwa eqondiwe” umsebenzisi ayilindele noma adinga ukuyizwa. Abacwaningi bahlola imodeli yabo ezigulini esezikhulile, futhi ababambiqhaza baba namandla kakhulu emini ngenxa yalokho. Ukwengeza, imodeli ingakwazi ukuthulisa iziguli futhi izipholise ukuze zilale ebusuku.

    Ngaleso sikhathi, i-voice synthesis nayo iya ngokuya isetshenziswa kumafilimu. Ukwenza isibonelo, ukwakha ukulandisa kwezwi okokwenziwa kochungechunge lwemibhalo ye-Netflix yango-2022, i-Andy Warhol Diaries, inkampani ekhiqiza izwi i-Resemble AI yasebenzisa imizuzu emi-3 nemizuzwana engu-12 yokuqoshwa kwezwi kwasekuqaleni kukaWarhol kusukela ngeminyaka yawo-1970s kanye nama-80s. Ubuchwepheshe bale nkampani buvumele izwi likaWarhol ukuthi liphinde lidalwe ukuze asho amazwi akhe kumadayari, enze idokhumentari enezingxenye eziyisithupha egxile empilweni yakhe.

    Ithimba lithathe okukhiqizwayo kwezwi lika-Warhol ku-AI futhi lenza izinguquko zomzwelo nephimbo. Baphinde bangeza ukungapheleli okufana nomuntu ngokubhekisela iziqeshana zomsindo zesinye isipikha. I-Resemble AI iphinda igcizelela ukuthi ngaphambi kwanoma iyiphi iphrojekthi yokuhlanganisa izwi noma i-synthesis, inkampani ihlala icela imvume kubanikazi bezwi noma abameleli babo bezomthetho. Ochungechungeni lwe-docu, inkampani ithole imvume ye-Andy Warhol Foundation.

    Imithelela yokuhlanganisa inkulumo

    Imithelela ebanzi yokuhlanganisa inkulumo ingase ihlanganise: 

    • Izinkampani zemidiya ezisebenzisa ukuhlanganiswa kwenkulumo ukuze zenze kabusha amazwi osaziwayo abashonile kumafilimu namadokhumentari. Kodwa-ke, ezinye izethameli zingase zikuthole kungenasimilo futhi kungenasisekelo.
    • Ukwanda kwezehlakalo zobugebengu be-inthanethi bokuhlanganisa izwi, ikakhulukazi embonini yezinsizakalo zezezimali.
    • Amafemu wezithombe ezibukhoma asebenzisa inkulumo yokwenziwa ukuze aphile imidwebo edumile nezibalo zomlando. Le nkonzo iheha ikakhulukazi kumamnyuziyamu kanye nomkhakha wezemfundo.
    • Ukuhlanganiswa kwenkulumo kusetshenziswa kumavidiyo angamanga ukuze kusabalaliswe inkulumo-ze futhi kubekwe abantu icala ngamanga, ikakhulukazi izintatheli nezishoshovu.
    • Amafemu amaningi okuqalisa agxile ekwenziweni kwezwi kanye nezinsizakalo zenkulumo zokwenziwa, okuhlanganisa osaziwayo namathonya abafuna ukuqasha amazwi abo kumabhrendi.
    • Ukubona okungokoqobo okuthuthukisiwe kubasizi ababonakalayo namageyimu asebenzisanayo ngokuhlanganiswa kwenkulumo okuthuthukisiwe, ukuthuthukisa ulwazi lomsebenzisi kodwa kuphakamisa ukukhathazeka ngokunamathela ngokomzwelo ku-AI.
    • Ukwamukelwa kokuhlanganiswa kwenkulumo kusevisi yamakhasimende ezenzakalelayo, ukwenza lula ukusebenza kodwa okungase kuholele ekususweni kwemisebenzi embonini yesikhungo sezingcingo.
    • Ama-ejensi kahulumeni asekela ukuhlanganiswa kwenkulumo ngezimemezelo zesevisi yomphakathi, okuvumela ukuxhumana kwezilimi eziningi kanye nephimbo elithile kodwa kudinga ukugadwa okucophelelayo ukuze kuvinjelwe ukusetshenziswa kabi noma ulwazi olungalungile.

    Imibuzo okufanele icatshangelwe

    • Yiziphi ezinye izinzuzo ezingaba khona zama-bot amaningi anomsindo womuntu?
    • Iyiphi enye indlela izigebengu ze-inthanethi ezingasebenzisa ngayo ukuhlanganiswa kwenkulumo?

    Izinkomba zokuqonda

    Izixhumanisi ezilandelayo ezidumile nezikhungo zibhekiselwe kulo mbono: