Samfuran RNN-Transducer
RNN-Transducer (RNN-T) ƙirar ƙirar magana ce mai yawo mai gudana wanda ke gyara babban rauni na CTC - rashin ikonsa na ƙirar dogaro tsakanin alamun fitarwa.
Dubawa
It powers much of the on-device 'live' speech recognition you use every day.
Zurfafa nutsewa
Hakanan Alex Graves (2012) ya gabatar da shi, RNN-Transducer ya haɗa abubuwa uku. Mai rikodin rikodin (cibiyar rubutun rubutu) tana aiwatar da firam ɗin sauti zuwa fasallan sauti. Cibiyar sadarwar tsinkaya tana aiki kamar ƙirar harshe, tana daidaita jerin alamun rubutu da aka fitar a baya. Ƙaramar hanyar sadarwa ta haɗin gwiwa sannan ta haɗa ra'ayi mai ɓoyewa na 'inda muke cikin sauti' tare da hangen nesa na cibiyar sadarwa na 'abin da muka faɗa zuwa yanzu' don nuna alama ta gaba akan ƙamus wanda ya haɗa da sarari. Ba kamar CTC ba, cibiyar sadarwar tsinkaya tana kawar da zato na sharadi, don haka RNN-T yana koyon ainihin rubutun kalmomi da tsarin kalmomi a ciki. Decoding yana tafiya da lattice na 2D na lokaci mai jiwuwa tare da alamun fitarwa, yana fitar da sarari don ci gaba ta hanyar sauti da alamun gaske don ci gaba ta hanyar rubutu - a zahiri yana goyan bayan fitowar yawo.
Fahimtar Fasaha
Asarar RNN-T, kamar na CTC, ta tattara duk ingantattun hanyoyin daidaitawa ta hanyar sake dawowa gaba, amma sama da grid mai girma biyu (matakan lokaci ta wuraren fitarwa) maimakon jeri ɗaya. Fitar da mara-fari yana tsayawa a firam ɗin sauti iri ɗaya kuma yana haɓaka alamar alamar; fitar da wani blank ci gaban lokaci. Wannan tsari na monotonic, hagu-zuwa-dama shine ainihin dalilin da yasa RNN-T ke gudana cikin tsafta tare da latency mai iyaka, sabanin cikakken hankali wanda zai iya leƙa a cikin duka furcin.
Dabarun Tasiri
Shiga ku isa
Yana inganta samun dama ta hanyar rubutu, ba da labari, da mu'amalar murya.
Kudin da kasafin kuɗi
Ƙungiyoyin kafofin watsa labaru na iya jigilar sauti mai gogewa cikin sauri tare da ƙaramin kasafin kuɗi.
Gudu da sikelin
Tsarin fuskantar abokin ciniki na iya aiwatar da hulɗar magana a mafi girman ma'auni.
Makomar Samfuran RNN-Transducer
RNN-T shine babban zaɓi don samarwa ASR yawo kuma yana ƙara amfani da masu rikodin Conformer maimakon LSTMs. Bincike ya mayar da hankali kan rage nauyin ƙwaƙwalwar ajiyarsa yayin horo, sarrafa latency don haka rubutun ya bayyana da sauri, da kuma daidaita 'saurin fitarwa'. Yi tsammanin ci gaba da haɗuwa tare da horarwa na kulawa da kai da masu fassara yaruka da yawa, tare da ƙaƙƙarfan tura na'ura yayin da ake ƙididdige tsinkaya da hanyoyin haɗin gwiwar haɗin gwiwa.
Aiwatar da Gaskiyar Duniya
Google's gane magana akan na'urar don ƙamus na Gboard da Pixel Recorder, yana aiki gabaɗaya ta layi
Taken kai tsaye wanda ke watsa kalmomi yayin da kuke magana maimakon jiran ku gama jumla
Mataimakan murya suna rubuta umarni tare da ƙarancin jinkiri yayin da kuke magana
Haɗuwa na ainihi da rubutun kira inda sakamakon sashe dole ne ya bayyana ci gaba
Hatsari & Tsare-tsare
Rashin amfani da murya da haɗarin kwaikwaya yana ƙaruwa lokacin da aka rasa izini.
Daidaituwa na iya faɗuwa cikin lafuzza, yaruka, ko mahalli masu hayaniya.
Ana iya kuskuren sauti na roba don ingantacciyar magana ba tare da bayyananniyar lakabi ba.
Taswirar Hanya
Sami tabbataccen izini don ɗaukar murya, cloning, da sake amfani.
Gwajin ingantattun masu magana daban-daban da yanayin baya.
Ƙayyade lokacin da dole ne ɗan adam ya duba ko ya amince da abubuwan da aka fitar.
Yi lakabin sauti na roba da kuma adana bayanan da aka tabbatar don yin lissafi.
Ci gaba da Bincike
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
Jagora na gaba
Rabuwar Hanyoyi Dual-Path RNN
Tambayoyin da ake yawan yi
What is RNN-Transducer Models?
RNN-Transducer (RNN-T) ƙirar ƙirar magana ce mai yawo mai gudana wanda ke gyara babban rauni na CTC - rashin ikonsa na ƙirar dogaro tsakanin alamun fitarwa. Yana ba da iko da yawa daga cikin na'urar 'rayuwa' fahimtar magana da kuke amfani da ita kowace rana.
Wanne iyakance CTC na RNN-Transducer ke magana musamman?
RNN-T yana ƙara hanyar sadarwar tsinkaya wacce sharuɗɗa akan alamun farko, cire zato na CTC cewa kowane fitarwa yana da zaman kansa idan aka ba da sautin.
Menene manyan abubuwa uku na RNN-Transducer?
RNN-T yana haɗa mai rikodin sauti tare da hanyar sadarwa mai sanyin rubutu, haɗaɗɗen ƙaramar hanyar sadarwa ta haɗin gwiwa wacce ke nuna alamar ta gaba.
Wace rawa cibiyar sadarwar hasashen ke takawa a cikin RNN-T?
Cibiyar sadarwar tsinkaya tana aiki azaman ƙirar harshe na ciki akan tarihin alamar fitarwa, yana ba da ainihin rubutun RNN-T da tsarin kalma.
Me yasa RNN-T ke goyan bayan raɗaɗin yawo a zahiri haka?
RNN-T yana tafiya da lattice na lokaci-lokaci na monotonic, don haka yana iya fitar da fitarwa yayin da sauti ya zo tare da latency mai iyaka maimakon jiran cikakken shirin.
A cikin ƙaddamarwa na RNN-T, menene fitar da alamar mara laifi?
A cikin lattice na RNN-T, babu komai yana ci gaba da axis na lokaci (matsa zuwa firam ɗin sauti na gaba), yayin da mara sarari yana ci gaba da axis na lakabin.