Amamodeli we-RNN-Transducer
I-RNN-Transducer (RNN-T) iyisakhiwo sokuqaphela inkulumo esilungele ukusakaza esilungisa ubuthakathaka obukhulu be-CTC — ukungakwazi kwayo ukwenza imodeli yokuncika phakathi kwamathokheni okukhiphayo.
Uhlolojikelele
It powers much of the on-device 'live' speech recognition you use every day.
I-Deep Dive
Futhi yethulwa ngu-Alex Graves (2012), iRNN-Transducer ihlanganisa izingxenye ezintathu. Isifaki khodi (inethiwekhi yokuloba) sicubungula ozimele bomsindo zibe izici ze-acoustic. Inethiwekhi yokuqagela isebenza njengemodeli yolimi, ibeka esimweni sokulandelana kwamathokheni ombhalo akhishwe ngaphambilini. Inethiwekhi encane ehlanganyelwe ibese ihlanganisa umbono wesifaki khodi wokuthi 'lapho sikhona kumsindo' nombono wenethiwekhi yokubikezela 'lokho esesikushilo kuze kube manje' ukuze uthole ithokheni elandelayo phezu kwamagama ahlanganisa nokungenalutho. Ngokungafani ne-CTC, inethiwekhi yokuqagela isusa ukuqagela kokuzimela okunemibandela, ngakho i-RNN-T ifunda isipelingi sangempela namaphethini wamagama ngaphakathi. Ukukhipha amakhodi kuhamba nge-lattice ye-2D yesikhathi somsindo uma kuqhathaniswa namathokheni okukhiphayo, kukhipha izikhala ukuze kuthuthukiswe ngomsindo namathokheni wangempela ukuze kuthuthuke ngombhalo - kusekela okukhiphayo okusakazwayo ngokwemvelo.
I-Technical Insight
Ukulahlekelwa kwe-RNN-T, njengama-CTC, kubala kuzo zonke izindlela zokuqondanisa ezivumelekile ngokuphindaphinda ukuya phambili, kodwa phezu kwegridi enezinhlangothi ezimbili (izinyathelo zesikhathi ngezindawo zokukhiphayo) esikhundleni sokulandelana okukodwa. Ukukhipha ukuhlala okungangenalutho kufreyimu yomsindo efanayo futhi kuthuthukisa inkomba yelebula; ukukhipha isikhathi sokuthuthuka esingenalutho. Lesi sakhiwo se-monotonic, kwesokunxele kuye kwesokudla yingakho i-RNN-T isakaza ngokuhlanzekile ngokubambezeleka okukhawulelwe, ngokungafani nokunaka okugcwele okungabheka kukho konke ukuphimisa.
I-Strategic Impact
Finyelela futhi ufinyelele
Ithuthukisa ukufinyeleleka ngokuloba, ukulandisa, nezixhumi ezibonakalayo zezwi.
Izindleko kanye nesabelomali
Amaqembu emidiya angathumela umsindo opholishiwe ngokushesha ngamabhajethi amancane.
Isivinini nesikali
Amasistimu abhekene nekhasimende angacubungula ukusebenzelana okukhulunyiwe ngesilinganiso esikhulu.
Ikusasa lamamodeli we-RNN-Transducer
I-RNN-T iyinketho ehamba phambili yokusakaza ukukhiqiza i-ASR futhi iya ngokuya isebenzisa izifaki khodi ze-Conformer esikhundleni sama-LSTM. Ucwaningo lugxile ekunciphiseni izindleko zenkumbulo ezisindayo phakathi nokuqeqeshwa, ukulawula ukubambezeleka kokukhishwa ukuze amagama-ncazo avele ngokushesha, kanye nokujwayelekile 'okuphuma ngokushesha'. Lindela ukuhlangana okuqhubekayo nokuziqeqesha kwangaphambili okuzigadile kanye nama-transducer ezilimi eziningi, kanye nokusetshenziswa okuqinile kudivayisi njengoba ukuqagela kanye namanethiwekhi ahlanganyelwe kulinganiswa futhi kusikwe.
Ukuqaliswa Komhlaba Wangempela
Google Ukubonwa kwenkulumo ekudivayisini kokubizelwa kwe-Gboard kanye ne-Pixel Recorder, esebenza ungaxhunyiwe ku-inthanethi ngokugcwele
Amagama-ncazo abukhoma asakaza amagama njengoba ukhuluma kunokulinda ukuthi uqedele umusho
Izilekeleli zezwi ezibhala imiyalo ngokubambezeleka okuphansi ngenkathi usakhuluma
Umhlangano wesikhathi sangempela kanye nokulotshwa kwekholi lapho imiphumela engaphelele kufanele ivele ngokuqhubekayo
Izingozi & Guardrails
Ukusetshenziswa kabi kwezwi kanye nezingozi zokuzenza ongeyena ziyanda uma imvume ingekho.
Ukunemba kungase kwehle kuzo zonke izinhlobo zokuphimisela, izilimi zesigodi, noma izindawo ezinomsindo.
Umsindo wokwenziwa ungenziwa iphutha njengenkulumo eyiqiniso ngaphandle kokulebula okucacile.
Ukuqalisa Umhlahlandlela
Thola imvume esobala yokuthwebula izwi, ukuhlanganisa, nokusebenzisa kabusha.
Ikhwalithi yokuhlola kuzo zonke izipikha nezimo zangemuva.
Chaza ukuthi kunini lapho umuntu kufanele abuyekeze noma agunyaze okuphumayo.
Lebula umsindo wokwenziwa futhi ugcine amarekhodi atholakalayo ukuze aziphendulele.
Qhubeka Uhlole
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the RNN-Transducer Models quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Umhlahlandlela olandelayo
Ukuhlukaniswa kwe-RNN ye-Dual-Path
Imibuzo evame ukubuzwa
What is RNN-Transducer Models?
I-RNN-Transducer (RNN-T) iyisakhiwo sokuqaphela inkulumo esilungele ukusakaza esilungisa ubuthakathaka obukhulu be-CTC — ukungakwazi kwayo ukwenza imodeli yokuncika phakathi kwamathokheni okukhiphayo. Inika amandla okuningi kokubonwa kwenkulumo 'okubukhoma' kudivayisi oyisebenzisa nsuku zonke.
Yimuphi umkhawulo we-CTC i-RNN-Transducer ekhuluma ngawo ngokuqondile?
I-RNN-T yengeza inethiwekhi yokuqagela ebeka izimo kumathokheni angaphambili, isusa ukucabanga kwe-CTC kokuthi okukhiphayo ngakunye kuzimele uma kubhekwa umsindo.
Yiziphi izingxenye ezintathu eziyinhloko ze-RNN-Transducer?
I-RNN-T ibhanqa isishumeki somsindo nenethiwekhi yokuqagela enesimo sombhalo, ehlanganiswe inethiwekhi encane ehlanganyelwe ehola ithokheni elandelayo.
Iyiphi indima edlalwa inethiwekhi yokubikezela ku-RNN-T?
Inethiwekhi yokuqagela isebenza njengemodeli yolimi lwangaphakathi phezu komlando wethokheni yokukhiphayo, inikeze i-RNN-T yangempela isipelingi namaphethini wamagama.
Kungani i-RNN-T isekela ukusakazwa kwe-latency ephansi ngokwemvelo?
I-RNN-T ihamba i-monotonic time-by-token lattice, ngakho-ke ingakwazi ukukhipha okukhiphayo njengoba umsindo ufika ubambezeleke kunomkhawulo kunokulinda isiqeshana esigcwele.
Ekuqopheni i-RNN-T, ukwenzani ukukhipha ithokheni elingenalutho?
Kulathisi we-RNN-T, okungenalutho kuqhubekisela phambili i-eksisi yesikhathi (iya kuzimele womsindo olandelayo), kuyilapho okungangenalutho kuqhubekisela phambili i-eksisi yelebula.