Inombolo yezinhlelo zokusebenza nokubaluleka kokuxhumana kwezwi kukhula ngokushesha
of technology

Inombolo yezinhlelo zokusebenza nokubaluleka kokuxhumana kwezwi kukhula ngokushesha

Umndeni waseMelika e-Portland, e-Oregon usanda kuthola ukuthi umsizi wezwi ka-Alex uqophe izingxoxo zawo zangasese futhi wazithumela kumngane. Umnikazi womuzi, obizwa ngokuthi uDanielle kwabezindaba, utshele abezindaba ukuthi "ngeke aphinde awuxhume lowo mshini ngoba akathenjwa."

Alexa, ehlinzekwe izipikha ze-Echo (1) namanye amagajethi emashumini ezigidi zamakhaya ase-US, iqala ukurekhoda lapho izwa igama layo noma "igama shayela" likhulunywa umsebenzisi. Lokhu kusho ukuthi noma igama elithi "Alexa" lishiwo esikhangisweni se-TV, idivayisi ingase iqale ukurekhoda. Yilokho kanye okwenzekile kuleli cala, kusho i-Amazon, umsabalalisi wehardware.

"Ingxoxo esele ihunyushwe yizwi njengomyalelo wokuthi kuthunyelwe umyalezo," kusho isitatimende senkampani. Ngesinye isikhathi, u-Alexa wabuza kakhulu: "Kubani?" Ukuqhubeka kwengxoxo yomndeni mayelana nokufakwa phansi ngokhuni oluqinile bekufanele kubonwe umshini njengento esohlwini lokuxhumana lwekhasimende.” Okungenani yilokho okucatshangwa yi-Amazon. Ngakho, ukuhunyushwa kuncishiswe ochungechungeni lwezingozi.

Nokho, ukukhathazeka kusekhona. Ngoba ngesizathu esithile, endlini lapho sisazizwa sikhululekile, kufanele singene uhlobo oluthile "lwemodi yezwi", sibuke esikushoyo, lokho okusakazwa yi-TV futhi, yiqiniso, ukuthi lesi sikhulumi esisha esifubeni kusho abadwebi . thina.

Nokho, Ngaphandle kokungapheleli kobuchwepheshe nokukhathazeka okuyimfihlo, ngokukhula kokuthandwa kwamadivayisi afana ne-Amazon Echo, abantu sebeqala ukujwayela umqondo wokusebenzelana namakhompyutha besebenzisa izwi labo..

Njengoba u-Werner Vogels, i-CTO yase-Amazon, aphawula ngesikhathi sakhe se-AWS re:Invent ngasekupheleni kuka-2017, ubuchwepheshe bukhawulele ikhono lethu lokusebenzelana namakhompyutha kuze kube manje. Sithayipha amagama angukhiye ku-Google sisebenzisa ikhibhodi, njengoba lena kuseyindlela evamile nelula yokufaka ulwazi emshinini.

Kusho uVogels. -

ezine ezinkulu

Lapho sisebenzisa injini yokusesha ye-Google ocingweni, mhlawumbe siqaphele uphawu lombhobho olunocingo lokukhuluma kudala. Lokhu I-Google manje (2), engasetshenziswa ukubiza umbuzo wosesho, ukufaka umlayezo ngezwi, njll. Eminyakeni yamuva, i-Google, i-Apple, ne-Amazon ithuthuke kakhulu. ubuchwepheshe bokubona izwi. Abasizi bezwi abanjengo-Alexa, uSiri, nomsizi weGoogle abagcini nje ngokuqopha izwi lakho, kodwa futhi bayakuqonda okushoyo kubo futhi baphendule imibuzo.

I-Google Now iyatholakala mahhala kubo bonke abasebenzisi be-Android. Uhlelo lokusebenza, ngokwesibonelo, lungakwazi ukusetha i-alamu, luhlole isimo sezulu futhi luhlole umzila ku-Google Maps. Isandiso sengxoxo sezifunda ze-Google Now Umsizi we-Google () - usizo lwe-virtual kumsebenzisi wemishini. Itholakala kakhulu kumaselula kanye namadivayisi ahlakaniphile asekhaya. Ngokungafani ne-Google Now, ingabamba iqhaza ekushintshisaneni ngezindlela ezimbili. Umsizi uphume okokuqala ngoMeyi 2016 njengengxenye yohlelo lokusebenza lokulayeza lwe-Google i-Allo, kanye nasesipikheni sezwi se-Google Home (3).

3. Ikhaya le-Google

Uhlelo lwe-IOS nalo lunomsizi walo obonakalayo, Siri, okuwuhlelo olufakwe nezinhlelo zokusebenza ze-Apple i-iOS, i-watchOS, i-tvOS homepod, ne-macOS. USiri uqale nge-iOS 5 kanye ne-iPhone 4s ngo-Okthoba 2011 engqungqutheleni ye-Let Talk iPhone.

Isofthiwe isekelwe ku-interface yengxoxo: ibona inkulumo yemvelo yomsebenzisi (nge-iOS 11 kungenzeka futhi ukufaka imiyalo ngesandla), iphendule imibuzo futhi iqedele imisebenzi. Sibonga ukwethulwa komshini wokufunda, umsizi ngokuhamba kwesikhathi ihlaziya okuthandwa nguwe umsebenzisi ukunikeza imiphumela ehambisana kakhulu nezincomo. I-Siri idinga uxhumano lwe-inthanethi njalo - imithombo eyinhloko yolwazi lapha i-Bing ne-Wolfram Alpha. I-iOS 10 yethule ukwesekwa kwezandiso zezinkampani zangaphandle.

Enye yezine ezinkulu UCortana. Iwumsizi womuntu siqu ohlakaniphile owakhiwe yi-Microsoft. Isekelwa ku-Windows 10, Windows 10 Mobile, Windows Phone 8.1, Xbox One, Skype, Microsoft Band, Microsoft Band 2, Android, and iOS platforms. I-Cortana yethulwa okokuqala engqungqutheleni yonjiniyela yeMicrosoft ngo-Ephreli 2014 eSan Francisco. Igama lohlelo livela egameni lomlingiswa ophuma ochungechungeni lwegeyimu ye-Halo. I-Cortana itholakala ngesiNgisi, isiNtaliyane, iSpanishi, isiFulentshi, isiJalimane, isiShayina nesiJapane.

Abasebenzisi bohlelo oselushiwo Alexa kufanele futhi bacabangele imikhawulo yolimi - umsizi wedijithali ukhuluma kuphela isiNgisi, isiJalimane, isiFulentshi nesiJapane.

I-Amazon Virtual Assistant yaqala ukusetshenziswa ku-Amazon Echo kanye nezipikha ezihlakaniphile ze-Amazon Echo Dot ezakhiwe yi-Amazon Lab126. Ivumela ukusebenzisana kwezwi, ukudlalwa komculo, ukudala uhlu lwezinto ozokwenziwa, ukusetha kwe-alamu, ukusakazwa kwe-podcast, ukudlalwa kwe-audiobook, nesimo sezulu sesikhathi sangempela, ithrafikhi, ezemidlalo, nolunye ulwazi lwezindaba njengezindaba (4). I-Alexa ingakwazi ukulawula amadivaysi amaningi ahlakaniphile ukuze idale isistimu ye-automation yasekhaya. Ingasetshenziswa futhi ukwenza ukuthenga okulula esitolo se-Amazon.

4. Yini Abasebenzisi Abayisebenzisela I-Echo (Ngokocwaningo)

Abasebenzisi bangathuthukisa ulwazi lwe-Alexa ngokufaka "amakhono" e-Alexa (), izici ezengeziwe ezithuthukiswe izinkampani zangaphandle, ezivame ukubizwa ngokuthi izinhlelo zokusebenza ezifana nesimo sezulu nezinhlelo zomsindo kwezinye izilungiselelo. Amadivayisi amaningi e-Alexa akuvumela ukuthi uvule umsizi wakho obonakalayo ngephasiwedi yokuvuka, ebizwa ngokuthi .

Namuhla, i-Amazon nakanjani ibusa imakethe yezikhulumi ezihlakaniphile (5). I-IBM, eyethula isevisi entsha ngo-March 2018, izama ukungena kwezine eziphezulu Umsizi kaWatson, eyenzelwe izinkampani ezifuna ukudala amasistimu azo abasizi ababonakalayo abanokulawula kwezwi. Iyini inzuzo yesixazululo se-IBM? Ngokusho kwabamele inkampani, okokuqala, emathubeni amakhulu kakhulu okwenza kube ngokwakho kanye nokuvikelwa kobumfihlo.

Okokuqala, umsizi we-Watson akaphawulwanga. Izinkampani zingakha izixazululo zazo kule nkundla futhi zizilebule ngomkhiqizo wazo.

Okwesibili, bangakwazi ukuqeqesha amasistimu abo okusiza besebenzisa amasethi abo edatha, i-IBM ethi yenza kube lula ukwengeza izici nemiyalo kuleyo sistimu kunobunye ubuchwepheshe be-VUI (isikhombimsebenzisi sezwi).

Okwesithathu, Umsizi we-Watson akayinikezi i-IBM ngolwazi mayelana nomsebenzisi - abathuthukisi bezixazululo endawenikazi bangagcina idatha ebalulekile kubo. Phakathi naleso sikhathi, noma ubani owakha amadivaysi, isibonelo nge-Alexa, kufanele aqaphele ukuthi idatha yabo ebalulekile izophela ku-Amazon.

Umsizi we-Watson usevele unokusebenzisa okumbalwa. Uhlelo lusetshenziswe, isibonelo, nguHarman, owakha umsizi wezwi wemoto yomqondo weMaserati (6). Esikhumulweni sezindiza saseMunich, umsizi we-IBM unika irobhothi amandla e-Pepper ukusiza abagibeli ukuhambahamba. Isibonelo sesithathu i-Chameleon Technologies, lapho ubuchwepheshe bezwi busetshenziswa kumitha yasekhaya ehlakaniphile.

6. Umsizi we-Watson emotweni yomqondo we-Maserati

Kufanelekile ukungeza ukuthi ubuchwepheshe obukhona lapha abubusha. Umsizi we-Watson uhlanganisa amandla okubethela emikhiqizo ekhona ye-IBM, Ingxoxo ye-Watson, ne-Watson Virtual Agent, kanye nama-API okuhlaziya ulimi nengxoxo.

I-Amazon ayiyena nje umholi kubuchwepheshe bezwi elihlakaniphile, kodwa iyenza ibe ibhizinisi eliqondile. Kodwa-ke, ezinye izinkampani ziye zazama ukuhlanganiswa kwe-Echo ngaphambili kakhulu. I-Sisense, inkampani esembonini ye-BI kanye nezibalo, yethula ukuhlanganiswa kwe-Echo ngoJulayi 2016. Ngokulandelayo, u-Roxy oqalayo wanquma ukudala isofthiwe yayo elawulwa ngezwi kanye ne-hardware yemboni yezokuvakasha. Ngasekuqaleni kwalo nyaka, i-Synqq yethule uhlelo lokusebenza lokuthatha amanothi olusebenzisa ukucubungula kwezwi nolimi lwemvelo ukuze lwengeze amanothi nokufakwa kwekhalenda ngaphandle kokuwathayipha kukhibhodi.

Wonke la mabhizinisi amancane anezifiso eziphezulu. Kodwa-ke, ngaphezu kwakho konke, bafunde ukuthi akuwona wonke umsebenzisi ofuna ukudlulisa idatha yakhe ku-Amazon, Google, Apple noma Microsoft, okungabadlali ababaluleke kakhulu ekwakheni izinkundla zokuxhumana ngezwi.

Abantu baseMelika bafuna ukuthenga

Ngo-2016, ukusesha ngezwi kubalele u-20% wakho konke ukusesha kweselula kwe-Google. Abantu abasebenzisa lobu buchwepheshe nsuku zonke bacaphuna ukusebenziseka kwabo nokwenza izinto eziningi phakathi kwezinzuzo zabo ezinkulu. (isibonelo, ikhono lokusebenzisa injini yokusesha ngenkathi ushayela imoto).

Abahlaziyi be-Visiongain balinganisela inani lamanje lemakethe labasizi bedijithali abahlakaniphile ku-$ 1,138. Ziningi izindlela ezinjalo. Ngokusho kukaGartner, ekupheleni kuka-2018 kakade U-30% wokusebenzelana kwethu ngobuchwepheshe kuzoba ngezingxoxo nezinhlelo zezwi.

Inkampani yocwaningo yaseBrithani i-IHS Markit ilinganisela ukuthi imakethe yabasizi bedijithali abasebenzisa amandla e-AI izofinyelela kumadivayisi ayizigidi eziyizinkulungwane ezine ekupheleni kwalo nyaka, futhi leso sibalo singase sikhuphuke sifinyelele ezigidini eziyizinkulungwane ezingu-4 ngo-2020.

Ngokwemibiko evela ku-eMarketer kanye ne-VoiceLabs, abantu baseMelika abayizigidi ezingama-2017 basebenzisa isilawuli sezwi okungenani kanye ngenyanga ngo-35,6. Lokhu kusho ukukhula cishe ngo-130% kunonyaka odlule. Imakethe yomsizi wedijithali iyodwa kulindeleke ukuthi ikhule ngo-2018% ngo-23. Lokhu kusho ukuthi uzobe usuzisebenzisa kakade. BaseMelika abayizigidi ezingama-60,5, okuzoholela emalini ephathekayo kubakhiqizi babo. I-RBC Capital Markets ilinganisela ukuthi i-Alexa interface izokhiqiza imali efika ku-$2020 billion yemali engenayo ye-Amazon ngo-10.

Geza, bhaka, hlanza!

Izixhumanisi zezwi ziya ngokuya zingena ngesibindi ezintweni zikagesi zasekhaya nasezimakethe zama-electronics abathengi. Lokhu kungase kubonakale kakade phakathi nombukiso we-IFA wangonyaka odlule 2017. Inkampani yaseMelika i-Neato Robotics yethula, isibonelo, isicoci se-robot vacuum esixhuma kwelinye lamapulatifomu amaningana asekhaya ahlakaniphile, kuhlanganise nesistimu ye-Amazon Echo. Ngokukhuluma nesipikha esihlakaniphile se-Echo, ungayalela umshini ukuthi uhlanze indlu yakho yonke ngezikhathi ezithile emini noma ebusuku.

Eminye imikhiqizo eyenziwe yasebenza ngezwi ikhonjisiwe kulo mdlalo, kusukela kuma-smart TV athengiswa ngaphansi kophawu lweToshiba yinkampani yaseTurkey i-Vestel kuya kwezingubo zokulala ezishisayo zenkampani yaseJalimane i-Beurer. Eziningi zalezi zisetshenziswa zikagesi zingenziwa zisebenze ukude kusetshenziswa ama-smartphone.

Kodwa-ke, ngokusho kwabamele i-Bosch, kusesekuseni kakhulu ukusho ukuthi yiziphi izinketho zomsizi wasekhaya ezizoba namandla. Ku-IFA 2017, iqembu lobuchwepheshe laseJalimane libonise imishini yokuwasha (7), ama-ovens nemishini yekhofi exhuma ku-Echo. I-Bosch iphinde ifune ukuthi amadivaysi ayo asebenzisane nezinkundla zezwi ze-Google ne-Apple ngokuzayo.

7. Umshini wokuwasha we-Bosch oxhuma ku-Amazon Echo

Izinkampani ezifana ne-Fujitsu, i-Sony ne-Panasonic zakha izixazululo zazo zomsizi wezwi ezisekelwe ku-AI. U-Sharp wengeza lobu buchwepheshe kumahhavini namarobhothi amancane angena emakethe. I-Nippon Telegraph & Telephone iqasha i-hardware nabenzi bamathoyizi ukuze bavumelane nesistimu yobuhlakani bokwenziwa elawulwa ngezwi.

Umqondo omdala. Ingabe isikhathi sakhe sesifikile?

Eqinisweni, umqondo we-Voice User Interface (VUI) usunamashumi eminyaka ukhona. Noma ubani owabuka i-Star Trek noma i-2001: I-Space Odyssey eminyakeni edlule cishe wayelindele ukuthi ngonyaka ka-2000 sonke sizolawula amakhompyutha ngamazwi ethu. Futhi, akubona ababhali bezinganekwane zesayensi kuphela ababone amandla alolu hlobo lwesixhumi esibonakalayo. Ngo-1986, abacwaningi bakwa-Nielsen babuza ochwepheshe be-IT ukuthi bacabanga ukuthi kungaba yini ushintsho olukhulu ekuxhumaneni komsebenzisi ngonyaka ka-2000. Bavame ukukhomba ekuthuthukisweni kwezixhumanisi zezwi.

Kunezizathu zokuthemba ikhambi elinjalo. Ukukhulumisana ngamazwi, phela, kuyindlela engokwemvelo kakhulu yokuthi abantu bashintshisane ngemicabango, ngakho ukuyisebenzisela ukusebenzisana nomshini womuntu kubonakala kuyisixazululo esingcono kakhulu kuze kube manje.

Enye ye-VUI yokuqala, ebizwa ibhokisi lezicathulo, yadalwa ekuqaleni kwawo-60s ngabakwa-IBM. Bekuyisiqalo sezinhlelo zanamuhla zokuzwa izwi. Kodwa-ke, ukuthuthukiswa kwamadivayisi we-VUI kwakunqunyelwe imingcele yamandla wekhompyutha. Ukuhlaziya nokuhumusha inkulumo yomuntu ngesikhathi sangempela kudinga umzamo omkhulu, futhi kwathatha iminyaka engaphezu kwamashumi amahlanu ukufika lapho kwenzeka ngempela.

Amadivayisi ane-interface yezwi aqala ukuvela ekukhiqizeni ngobuningi maphakathi nawo-90s, kodwa awazange athole ukuthandwa. Ucingo lokuqala olunokulawulwa kwezwi (ukudayela) kwaba Philips Sparkyakhululwa ngo-1996. Kodwa-ke, le divayisi entsha nesebenziseka kalula yayingenayo imikhawulo yezobuchwepheshe.

Amanye amafoni afakwe izinhlobo zezwi (akhiwe izinkampani ezifana ne-RIM, i-Samsung noma i-Motorola) ahlala efika emakethe, okuvumela abasebenzisi ukuthi bashayele ngezwi noma bathumele imilayezo yombhalo. Nokho, yonke yayidinga ukubamba ngekhanda imiyalo ethile nokuyisho ngendlela ephoqelelwe, yokwenziwa, evunyelaniswa namakhono emishini yangaleso sikhathi. Lokhu kudale inani elikhulu lamaphutha, okwaholela ekunganeliseki komsebenzisi.

Nokho, manje singena enkathini entsha yokwenza ikhompuyutha, lapho intuthuko ekufundeni komshini nobuhlakani bokwenziwa ivula amandla engxoxo njengendlela entsha yokuxhumana nobuchwepheshe (8). Inani lamadivayisi asekela ukusebenzisana kwezwi libe yinto ebalulekile eye yaba nomthelela omkhulu ekuthuthukisweni kwe-VUI. Namuhla, cishe u-1/3 wabantu bomhlaba sebevele bengabanikazi bama-smartphones angasetshenziselwa lolu hlobo lokuziphatha. Kubonakala sengathi abasebenzisi abaningi ekugcineni sebekulungele ukujwayelanisa izixhumanisi zezwi labo.

8. Umlando wesimanje wokuthuthukiswa kwesixhumi esibonakalayo sezwi

Nokho, ngaphambi kokuba sikhulume ngokukhululekile nekhompyutha, njengoba kwenza amaqhawe e-A Space Odyssey, kufanele sinqobe izinkinga ezimbalwa. Imishini namanje ayikabi muhle kakhulu ekuphatheni ama-nuances wolimi. Ngaphandle kwalokho abantu abaningi basazizwa bengakhululekile ukunikeza imiyalo yezwi enjinini yokusesha.

Izibalo zibonisa ukuthi izilekeleli zezwi zisetshenziswa kakhulu ekhaya noma phakathi kwabangane abaseduze. Akekho noyedwa kulabo okwaxoxwa naye owavuma ukuthi basebenzisa ukusesha ngezwi ezindaweni zomphakathi. Kodwa-ke, lokhu kuvinjelwa kungenzeka kunyamalale ngokusabalala kwalobu buchwepheshe.

umbuzo onzima ngokobuchwepheshe

Inkinga amasistimu (ASR) abhekene nayo ikhipha idatha ewusizo esignali yenkulumo futhi ihlobanise negama elithile elinencazelo ethile kumuntu. Imisindo ekhiqizwayo ihlukile isikhathi ngasinye.

Ukuhlukahluka kwesignali yenkulumo impahla yayo yemvelo, sibonga ngayo, ngokwesibonelo, siqaphela isisho noma iphimbo. Ingxenye ngayinye yesistimu yokuqaphela inkulumo inomsebenzi othile. Ngokusekelwe kusignali ecutshunguliwe kanye nemingcele yayo, imodeli ye-acoustic iyakhiwa, ehlotshaniswa nemodeli yolimi. Uhlelo lokuqaphela lungasebenza ngesisekelo senani elincane noma elikhulu lamaphethini, elinquma ubukhulu besilulumagama elisebenza ngaso. Bangase babe izichazamazwi ezincane endabeni yezinhlelo eziqaphela amagama noma imiyalo ngayinye, kanye database ezinkulu equkethe okulingana nesethi yolimi futhi kucatshangelwa imodeli yolimi (uhlelo lolimi).

Izinkinga ezibhekene nokuxhumana kwezwi kwasekuqaleni qonda inkulumo ngendlela efanele, isibonelo, lapho, ngokwesibonelo, lonke ukulandelana kohlelo lolimi kuvame ukushiywa, amaphutha olimi nefonetiki, amaphutha, okushiywa ngaphandle, ukungalungi kwenkulumo, amagama afanayo, ukuphindaphinda okungenasizathu, njll. Zonke lezi zinhlelo ze-ACP kufanele zisebenze ngokushesha nangokuthembekile. Okungenani lezo yizinto ezilindelwe.

Umthombo wobunzima futhi amasiginali we-acoustic ngaphandle kwenkulumo eyaziwayo efaka okokufaka kwesistimu yokuqaphela, i.e. zonke izinhlobo ukuphazamiseka nomsindo. Esimweni esilula, uyazidinga hlunga ngaphandle. Lo msebenzi ubonakala ujwayelekile futhi ulula - emva kwakho konke, amasiginali ahlukahlukene ayahlungwa futhi wonke unjiniyela we-elekthronikhi uyazi ukuthi enzeni esimweni esinjalo. Kodwa-ke, lokhu kufanele kwenziwe ngokucophelela nangokucophelela uma umphumela wokuqashelwa kwenkulumo uhlangabezana nesikulindele.

Ukuhlunga okusetshenziswa njengamanje kwenza kube nokwenzeka ukususa, kanye nesignali yenkulumo, umsindo wangaphandle othathwe imakrofoni kanye nezakhiwo zangaphakathi zesignali yenkulumo ngokwayo, okwenza kube nzima ukuyibona. Kodwa-ke, inkinga yobuchwepheshe eyinkimbinkimbi kakhulu iphakama lapho ukuphazamiseka kwesignali yenkulumo ehlaziywe ... enye isignali yenkulumo, okungukuthi, isibonelo, izingxoxo ezinkulu ezizungezile. Lo mbuzo waziwa ezincwadini ngokuthi lo okuthiwa . Lokhu kakade kudinga ukusetshenziswa kwezindlela eziyinkimbinkimbi, okuthiwa. i-deconvolution (eqaqa) uphawu.

Izinkinga zokuqashelwa kwenkulumo azigcini lapho. Kuyafaneleka ukuqaphela ukuthi inkulumo iphethe izinhlobo eziningi zolwazi. Izwi lomuntu liphakamisa ubulili, ubudala, izinhlamvu ezihlukene zomnikazi noma isimo sempilo yakhe. Kunomnyango obanzi wobunjiniyela bezinto eziphilayo obhekene nokuxilonga izifo ezihlukahlukene ngokusekelwe kusici sezenzakalo ze-acoustic ezitholakala kusignali yenkulumo.

Kukhona futhi izinhlelo zokusebenza lapho inhloso eyinhloko yokuhlaziya i-acoustic yesignali yenkulumo kuwukuhlonza isikhulumi noma ukuqinisekisa ukuthi singubani (izwi esikhundleni sokhiye, iphasiwedi noma ikhodi ye-PUK). Lokhu kungabaluleka, ikakhulukazi kubuchwepheshe bokwakha obuhlakaniphile.

Ingxenye yokuqala yesistimu yokuqaphela inkulumo ithi imakrofoni. Kodwa-ke, isignali ethathwe imakrofoni ngokuvamile ihlala ingasebenzi. Ucwaningo lubonisa ukuthi ukwakheka nokuhamba kwegagasi lomsindo kuyehluka kakhulu kuye ngomuntu, isivinini sokukhuluma, futhi ngokwengxenye isimo somuntu okhuluma naye - kuyilapho ngokwezinga elincane abonisa okuqukethwe yimiyalelo ekhulunywayo.

Ngakho-ke, isignali kufanele icutshungulwe ngendlela efanele. Imisindo yesimanjemanje, ifonetiki nesayensi yekhompiyutha ndawonye kunikeza isethi ecebile yamathuluzi angasetshenziswa ukucubungula, ukuhlaziya, ukubona nokuqonda isignali yenkulumo. I-spectrum eguqukayo yesignali, okuthiwa ama-spectrogram ashukumisayo. Kulula ukuzithola, futhi inkulumo ethulwa ngendlela ye-spectrogram eguqukayo kulula ukuyibona kusetshenziswa amasu afana nalawo asetshenziswa ekuboneni isithombe.

Izingxenye ezilula zenkulumo (isibonelo, imiyalo) zingabonwa ngokufana okulula kwama-spectrograms wonke. Isibonelo, isichazamazwi sikamakhalekhukhwini esicushwe yizwi siqukethe amagama nemishwana embalwa nje kuphela ukuya kwamakhulu ambalwa, evamise ukupakishwa ngaphambili ukuze ibonakale kalula nangempumelelo. Lokhu kwanele kwimisebenzi yokulawula elula, kodwa ikhawulela kakhulu uhlelo lokusebenza lulonke. Amasistimu akhiwe ngokuvumelana nohlelo, njengomthetho, asekela izikhulumi ezithile kuphela lapho amazwi aqeqeshwe ngokukhethekile. Ngakho-ke uma kukhona omusha ofuna ukusebenzisa izwi lakhe ukuze alawule uhlelo, cishe ngeke amukelwe.

Umphumela walo msebenzi ubizwa ngokuthi I-spectrogram engu-2-W, okungukuthi, i-spectrum enezinhlangothi ezimbili. Kukhona omunye umsebenzi kule block okufanele ukunakwa - ukuhlukaniswa. Ngokuvamile, sikhuluma ngokuhlukanisa isignali yenkulumo eqhubekayo ibe izingxenye ezingabonwa ngokuhlukana. Kusuka kulokhu kuxilongwa komuntu ngamunye kuphela lapho kwakhiwa khona ukuqashelwa okuphelele. Le nqubo iyadingeka ngoba akunakwenzeka ukukhomba inkulumo ende neyinkimbinkimbi ngesikhathi esisodwa. Imiqulu ephelele isivele ibhaliwe mayelana nokuthi yiziphi izingxenye okufanele zihlukaniseke kusignali yenkulumo, ngakho-ke ngeke sinqume manje ukuthi izingxenye ezihlukanisiwe kufanele zibe amafonimu (okulingana nemisindo), amalunga, noma mhlawumbe ama-allofoni.

Inqubo yokuqaphela okuzenzakalelayo njalo ibhekisela kwezinye izici zezinto. Amakhulu amasethi amapharamitha ahlukene ahlolelwe isignali yenkulumo. Isignali yenkulumo ine ihlukaniswe yaba ozimele abaziwayo nokuba izici ezikhethiwelapho lezi zinhlaka zethulwa khona enqubweni yokuqashelwa, singakwazi ukwenza (kuhlaka ngalunye ngokwehlukana) ukuhlukaniswa, i.e. ukwabela isihlonzi kuhlaka, oluzoyimela ngokuzayo.

Isigaba esilandelayo ukuhlanganisa amafreyimu abe amagama ahlukene - ngokuvamile esekelwe kulokho okuthiwa. imodeli yamamodeli we-Markov angacacile (HMM-). Bese kulandela ukuhlangana kwamagama qedela imisho.

Manje singabuyela ohlelweni lwe-Alexa okwesikhashana. Isibonelo sakhe sibonisa inqubo yezigaba eziningi zomshini "ukuqonda" komuntu - ngokunembile: umyalo onikezwe nguye noma umbuzo obuziwe.

Ukuqonda amagama, ukuqonda incazelo, nokuqonda inhloso yomsebenzisi kuyizinto ezihluke ngokuphelele.

Ngakho-ke, isinyathelo esilandelayo umsebenzi we-NLP module (), umsebenzi wawo ukuqashelwa kwenhloso yomsebenzisi, i.e. incazelo yomyalelo/umbuzo esimweni lapho ushiwo khona. Uma inhloso ikhonjwa, ke ukwabelwa lokho okubizwa ngamakhono namakhono, okungukuthi isici esithile esisekelwa umsizi ohlakaniphile. Endabeni yombuzo mayelana nesimo sezulu, imithombo yedatha yesimo sezulu ibizwa ngokuthi, okusamele icutshungulwe ibe yinkulumo (TTS - mechanism). Ngenxa yalokho, umsebenzisi uzwa impendulo yombuzo obuziwe.

Izwi? Ubuciko bezithombe? Noma mhlawumbe kokubili?

Amasistimu okusebenzisana amaningi esimanje asekelwe kumuntu obizwa ngokuthi i-graphical interface yomsebenzisi (i-graphical interface). Ngeshwa, i-GUI akuyona indlela esobala kakhulu yokusebenzisana nomkhiqizo wedijithali. Lokhu kudinga ukuthi abasebenzisi baqale bafunde ukusebenzisa isixhumi esibonakalayo futhi bakhumbule lolu lwazi ngokusebenzisana ngakunye okulandelayo. Ezimweni eziningi, izwi lilula kakhulu, ngoba ungakwazi ukuxhumana ne-VUI ngokukhuluma nedivayisi. Ukuxhumana okungaphoqi abasebenzisi ukuthi babambe ngekhanda futhi babambe ngekhanda imiyalo ethile noma izindlela zokusebenzisana kubangela izinkinga ezimbalwa.

Vele, ukunwetshwa kwe-VUI akusho ukuyeka ukuxhumana okungokwesiko okuthe xaxa - kunalokho, kuzotholakala izixhumanisi ezihlanganisiwe ezihlanganisa izindlela ezimbalwa zokuxhumana.

Isixhumi esibonakalayo asifanele yonke imisebenzi emongweni weselula. Ngalo, sizobiza umngane oshayela imoto, futhi simthumele ngisho ne-SMS, kodwa ukuhlola ukudluliselwa kwakamuva kungase kube nzima kakhulu - ngenxa yolwazi oludluliselwe ohlelweni () futhi olukhiqizwa uhlelo (uhlelo). Njengoba u-Rachel Hinman ephakamisa encwadini yakhe ethi Mobile Frontier, ukusebenzisa i-VUI kuba yimpumelelo kakhulu lapho wenza imisebenzi lapho inani lokufakwayo nolwazi oluphumayo lilincane.

I-smartphone exhunywe ku-inthanethi ilula kodwa futhi ayiphazamiseki (9). Ngaso sonke isikhathi uma umsebenzisi efuna ukuthenga okuthile noma ukusebenzisa isevisi entsha, kufanele alande olunye uhlelo lokusebenza futhi adale i-akhawunti entsha. Inkambu yokusetshenziswa nokuthuthukiswa kwezokuxhumana ngezwi idalwe lapha. Esikhundleni sokuphoqa abasebenzisi ukuthi bafake izinhlelo zokusebenza eziningi ezahlukene noma benze ama-akhawunti ahlukene esevisi ngayinye, ochwepheshe bathi i-VUI izosusa umthwalo wale misebenzi enzima iwuyise kumsizi wezwi onikwe amandla yi-AI. Kuyoba lula ngaye ukwenza imisebenzi ekhandlayo. Sizomnika imiyalo kuphela.

9. Ukuxhumana kwezwi nge-smart phone

Namuhla, okungaphezu nje kwefoni nekhompyutha kuxhumeke ku-inthanethi. Ama-smart thermostat, izibani, amaketela namanye amadivaysi amaningi ahlanganiswe ne-IoT nawo axhunywe kunethiwekhi (10). Ngakho-ke, kunamadivayisi angenawaya ezisizungezile agcwalisa izimpilo zethu, kodwa akuwona wonke angena ngokwemvelo esibonakalayo esibonakalayo somsebenzisi. Ukusebenzisa i-VUI kuzokusiza ukuthi uwahlanganise kalula endaweni yethu.

10. Ukuxhumana kwezwi ne-inthanethi yezinto

Ukudala i-interface yomsebenzisi yezwi maduze kuzoba ikhono lomklami elibalulekile. Lokhu kuyinkinga yangempela - isidingo sokuqalisa izinhlelo zezwi sizokukhuthaza ukuthi ugxile kakhulu ekwakhiweni okusebenzayo, okungukuthi, ukuzama ukuqonda izinhloso zokuqala zomsebenzisi, ukulindela izidingo zabo kanye nalokho abakulindele kuzo zonke izigaba zengxoxo.

I-Voice iyindlela ephumelelayo yokufaka idatha—ivumela abasebenzisi ukuthi bakhiphe ngokushesha imiyalo ohlelweni ngemibandela yabo. Ngakolunye uhlangothi, isikrini sinikeza indlela ephumelelayo yokubonisa ulwazi: sivumela amasistimu ukuthi abonise inani elikhulu lolwazi ngesikhathi esifanayo, ukunciphisa umthwalo kwimemori yabasebenzisi. Kunengqondo ukuthi ukuzihlanganisa zibe isimiso esisodwa kuzwakala kukhuthaza.

Izipikha ezihlakaniphile ezifana ne-Amazon Echo ne-Google Home azinikezi nhlobo isibonisi esibonakalayo. Ukuthuthukisa ngokuphawulekayo ukunemba kokuqashelwa kwezwi emabangeni aphakathi, kuvumela ukusebenza kwezandla, okubuye kukhulise ukuguquguquka kwabo nokusebenza kahle - bayafiseleka ngisho nakubasebenzisi asebevele banama-smartphones ngokulawula izwi. Nokho, ukuntuleka kwesikrini kuwumkhawulo omkhulu.

Amabhiphu kuphela angasetshenziswa ukwazisa abasebenzisi ngemiyalo engenzeka, futhi ukufunda okukhiphayo ngokuzwakalayo kuba yisicefe ngaphandle kwemisebenzi eyisisekelo kakhulu. Ukusetha isibali sikhathi ngomyalo wezwi ngenkathi upheka kuhle, kodwa ukwenza ubuze ukuthi singakanani isikhathi esisele akudingekile. Ukuthola isibikezelo sezulu esivamile kuba uvivinyo lwenkumbulo kumsebenzisi, okufanele alalele futhi amunce uchungechunge lwamaqiniso isonto lonke, kunokuba awathathe esikrinini ngokubuka nje.

Abaqambi sebenayo kakade isixazululo se-hybrid, I-Echo Show (11), eyengeze isikrini kusipikha esihlakaniphile se-Echo. Lokhu kukhulisa kakhulu ukusebenza kwemishini. Kodwa-ke, i-Echo Show isenamandla amancane okwenza imisebenzi eyisisekelo osekunesikhathi eside ikhona kuma-smartphones kanye namathebulethi. Ayikwazi (okwamanje) ukusebenzisa iwebhu, ukubonisa ukubuyekezwa, noma ukubonisa okuqukethwe kwekalishi lokuthenga lase-Amazon, isibonelo.

Isibonisi esibonakalayo siyindlela ephumelela kakhulu yokunikeza abantu ingcebo yolwazi kunomsindo nje. Ukuklama ngokubaluleka kwezwi kungathuthukisa kakhulu ukusebenzisana kwezwi, kodwa ngokuhamba kwesikhathi, ngokunganaki ukusebenzisa imenyu yokubuka ngenjongo yokusebenzelana kuzofana nokulwa ubophe isandla esisodwa ngemuva kwakho. Ngenxa yobunkimbinkimbi obuzayo bezwi elihlakaniphile elisuka ekupheleni liye ekupheleni nezibonisi, onjiniyela kufanele bacabangele ngokungathi sína indlela eyingxube yezixhumanisi.

Ukwandisa ukusebenza kahle kanye nesivinini sokukhiqizwa kwenkulumo kanye nezinhlelo zokuqashelwa kwenze kwaba nokwenzeka ukuzisebenzisa ezinhlelweni ezinjalo nasezindaweni ezifana, njengesibonelo:

• ezempi (imiyalo yezwi ezindizeni noma ezindizeni ezinophephela emhlane, isibonelo, F16 VISTA),

• ukuloba okuzenzakalelayo kombhalo (inkulumo iye embhalweni),

• amasistimu olwazi asebenzisanayo (Inkulumo Eyinhloko, izingosi zezwi),

• amadivaysi eselula (amafoni, ama-smartphone, amaphilisi),

• amarobhothi (Cleverbot - ASR amasistimu ahlanganiswe nobuhlakani bokwenziwa),

• ezezimoto (ukulawula okungaphathwa ngesandla kwezingxenye zemoto, njengeBlue & Me),

• izinhlelo zokusebenza zasekhaya (amasistimu asekhaya ahlakaniphile).

Qaphela ukuphepha!

Izimoto, izinto zikagesi zasendlini, ukushisisa/ukupholisa kanye nezinhlelo zokuphepha zasekhaya, kanye nenqwaba yempahla yasendlini isiqala ukusebenzisa izixhumanisi zezwi, ngokuvamile ezisekelwe ku-AI. Kulesi sigaba, idatha etholwe ezigidini zezingxoxo nemishini ithunyelwa amafu ekhompyutha. Kuyacaca ukuthi abakhangisi banentshisekelo kubo. Futhi hhayi bona kuphela.

Umbiko wakamuva ovela kochwepheshe bezokuphepha be-Symantec uncoma ukuthi abasebenzisi bomyalo wezwi bangalawuli izici zokuphepha ezifana nezingidi zeminyango, ingasaphathwa eyezinhlelo zokuphepha zasekhaya. Okufanayo kuya ekugcineni amaphasiwedi noma ulwazi oluyimfihlo. Ukuvikeleka kobuhlakani bokwenziwa kanye nemikhiqizo ehlakaniphile akukakacutshungulwa ngokwanele.

Lapho amadivayisi ekhaya lonke elalela wonke amagama, ubungozi bokugetshengwa kwesistimu nokusetshenziswa kabi kuba yinto ekhathazayo enkulu. Uma umhlaseli efinyelela kunethiwekhi yendawo noma amakheli ayo e-imeyili ahlobene, izilungiselelo zedivayisi ehlakaniphile zingashintshwa noma zisethwe kabusha zibe izilungiselelo zasembonini, okuzoholela ekulahlekelweni kolwazi olubalulekile kanye nokususwa komlando womsebenzisi.

Ngamanye amazwi, ochwepheshe bezokuphepha besaba ukuthi i-AI eshayelwa ngezwi kanye ne-VUI abakahlakaniphi ngokwanele ukuze basivikele ezinsongweni ezingaba khona futhi bagcine imilomo yethu ivaliwe lapho umuntu esingamazi ecela okuthile.

Engeza amazwana