Ukutholwa kokuchofoza kokhiye ngomsindo kufinyelelwa ngokubeka idivayisi yokulalela eduze nokuhlosiwe
Ngaso sonke isikhathi lapho ngihlatshwa umxhwele izindlela ezitholakele kanye/noma ezithuthukisiwe zombili ukuthola ulwazi kanye nokufinyelela esigabeni se-X, gqekeza idivayisi ye-X futhi ukuthi kuze kube manje lezo ezisangihlaba umxhwele kakhulu yilezo ezisekelwe imisindo ekhiqizwe abalandeli be-CPU ukuze bathole ulwazi, kanye nokusetshenziswa kokubona ngezindonga, phakathi kokunye okuningi.
Yingakho, ngokwami, ngithanda ngempela ukwabelana ngalolu hlobo lwezindaba lapha kubhulogi futhi esimweni Namuhla ngizokwabelana ngezindaba mayelana ukuthuthukiswa kwe- indlela ukuthi wakha ithimba labacwaningi aba Isekelwe ekutholeni ulwazi olufakwe kukhibhodi 95% kunembe uma uhlaziya umsindo wezinkinobho ezirekhodwe i-smartphone eseduze noma othathwe imakrofoni eseduze.
Ukunemba kokuthola okokufaka okuqinisekisiwe idlula zonke izindlela zokuhlaziya ama-acoustic uhlamvu ngophawu ngalunye eyaziwa ngaphambilini Abasebenzisi imodeli yolimi. Indlela ehlongozwayo ingasetshenziswa, isibonelo, ukunquma amagama ayimfihlo afakiwe noma imilayezo ethayiphiwe, esimweni lapho umhlaseli ebeke khona i-smartphone yakhe eduze kwesisulu noma ethola umsindo orekhodiwe ngenkathi efaka imininingwane eyimfihlo (isibonelo, lapho phakathi nokuxhumana isisulu singena ngephasiwedi yezinye izinhlelo zolwazi).
Ngokuthuthuka kwakamuva ekufundeni okujulile, ukuba khona kwendawo yonke yemakrofoni, kanye nokwanda kwezinsizakalo ze-inthanethi ngemishini yomuntu siqu, ukuhlaselwa kwesiteshi esiseceleni kwe-acoustic kuveza usongo olukhulu kumakhibhodi kunangaphambili.
Okokufaka kudalwe kabusha kusetshenziswa isihlukanisi esisuselwe kumodeli yokufunda yomshini ecabangela izici zomsindo nezinga levolumu lapho kucindezelwa okhiye abahlukene.
Kushiwo lokho ukwenza ukuhlasela, ukuqeqeshwa kokuqala kwemodeli kuyadingeka, okudinga ukufaniswa komsindo wokufakwayo nolwazi mayelana nezinkinobho ezicindezelwayo. Ngaphansi kwezimo ezikahle, imodeli ingaqeqeshwa kusetshenziswa uhlelo olungayilungele ikhompuyutha efakwe kukhompuyutha ehlaselwe, okwenza kube nokwenzeka ukuqopha umsindo ovela kumakrofoni kanyekanye futhi ubambe ama-keystrokes.
Esimeni esingokoqobo kakhudlwana, idatha edingekayo ukuze kuqeqeshwe imodeli ingaqoqwa ngokufanisa imilayezo yombhalo yokufaka nomsindo ovela kusethi erekhodiwe njengomphumela wenkomfa yevidiyo. Ukunemba kokutholwa kokokufaka lapho uqeqesha imodeli esuselwe ku-Zoom kanye nokuhlaziywa kokufakwayo kwenkomfa ngevidiyo ye-Skype kwehla kancane kuye ku-93% no-91,7%, ngokulandelana.
Esivivinyweni sokuqeqesha imodeli yokufunda yomshini kusetshenziswa umsindo ovela kunkomfa ye-Zoom, ukhiye ngamunye kwabangu-36 (0-9, a-z) kukhibhodi ucindezelwe izikhathi ezingu-25 zilandelana ngeminwe ehlukene nangamandla ahlukene.
Idatha emayelana nomsindo womshini ngamunye yaguqulwa yaba isithombe nge-spectrogram ekhombisa ukuguquka kwemvamisa kanye nokuphakama komsindo ngokuhamba kwesikhathi
ama-spectrogram idluliselwe ukuqeqeshelwa ukulinganisa ngokusekelwe kumodeli ye-CoAtNet, esetshenziselwa ukuhlukaniswa kwezithombe ezinhlelweni zombono wokwenziwa. Okusho ukuthi, ngesikhathi sokuqeqesha isithombe siqhathaniswa ne-spectrogram ye-keystroke ngayinye enegama lokhiye. Ukuze unqume okhiye abacindezelwe umsindo, imodeli ye-CoAtNet ibuyisela ukhiye okungenzeka kakhulu ngokusekelwe ku-spectrogram edlulisiwe, efana nokubuyisela ilebula okungenzeka kakhulu uma ibona izinto ngesithombe sazo.
Ngokuzayo, abacwaningi bahlose ukuhlola ithuba lokudala kabusha okokufaka kwekhibhodi ngokurekhoda umsindo wezipikha ezihlakaniphile futhi, ukuze bathuthukise ukunemba kokunquma umbhalo wokufakwayo, basebenzise imodeli yolimi ehlukanisa okokufakayo kumongo wamagama aphelele.
Okokugcina uma unentshisekelo yokwazi kabanzi ngakho, ungabheka imininingwane kufayela le- isixhumanisi esilandelayo.