RNN-Modèlu Tekkikatu
RNN-Transducer (RNN-T) ab architecture buy xàmmee kàddu la buy saafara ñakk kattan gu gëna mag ci CTC - ñàkka mëna modele dependence yi ci digganté token yiy génne.
Résumé
It powers much of the on-device 'live' speech recognition you use every day.
Plongeur bu xóot
Alex Graves (2012) moo ko njëkka dugal, tekkikatu RNN dafa boole ñatti mbir. Encodeur bi (reso transkripsioŋ bi) dafay soppi kadre audio yi ñu nekk màndarga akustik. Reseau biy wax luy am dafay melni modelu làkk, di aju ci toppalante token yiñ njëkka joxe. Benn reso bu ndaw buñ boole dafay boole gis-gis encoder bi ci 'fi ñu nekk ci audio bi' ak gis-gis reso biy wax luy waaja am ci 'li ñu wax ba leegi' ngir dugal token bi ci topp ci kaw vocabulaire bu am lu amul dara. RNN-T wuute na ak CTC ndax reso biy wax luy am dafay dindi xalaatu moomel sa bopp ci anam wu jaar yoon, moo tax RNN-T dafay jàng ortograafi ak motif baat yi ci biir. Decoding dafay dox ci 2D lattice audio-time ak jetons-output, di génne ay blank ngir awaase ci audio ak jetons dëgg ngir awaase ci bind - di jàppale streaming output.
Gis-gis xarala
RNN-T's perte, melni CTC's, dafay boole ci yooni yoon yépp jaaraleko ci recursion jëm kanam-dellu ginaaw, waaye ci kaw griy bu am ñaari dimension (jotu jéego ci position yi ñuy génne) du benn yoon. Sooy génne lu amul dara, dafay des ci benn kadre audio bi, ba noppi gëna yokk limu etiket bi; génne ab bërëb bu amul dara. Bii xeetu monotonic, cammoy-ci-ndeyjoor moo waral RNN-T di stream bu baax ak latency bu yam, wuute ak bàyyi xel bu mat sëkk bi mëna xool wax ji yépp.
njeextalu pexe
Dugg ak yegg
Dafay gëna yombal jëfandikoo gi jaaraleko ci transkripsioŋ, nettali ak interfaasu baat.
Njëgg ak budget
Ekipu mejaa yi mën nañu yónnee audio bu leer ci anam wu gëna gaaw te seen xaalis gëna néew.
Gaawaay ak yaatuwaay
Sistem yiy jàkkarloo ak kiliyaan bi mën nañu def waxtaan ci anam wu gëna yaatu.
Ëlëgu xeetu RNN-Transducer
RNN-T mooy tanneef bi gëna am solo ngir defar ASR te dafay gëna jëfandikoo encodeur Conformer ci palaasu LSTMs. Gëstu bi dafay sëgg ci wàññi njëgu mémoire bu diis bi ci diiru tàggat yaram, di saytu latency emision suko defee caption yi feeñ ci saasi, ak 'fast emit' regularisation. Xaarandil wéyal ndaje ak tàggat yaram bu ñuy saytu seen bopp ak transducer yu bari làkk, boole ci gëna dëgër ci aparey bi ñuy wax luy waaja am ak reso yuñ boole dañuy xayma ak dagg.
Doxal ci àdduna dëgg
Google's xàmmee kàddu yi ci aparey bi ngir dikte Gboard ak enregistreur pixel, di dox lëmm
Kaption en direct buy streaming kàddu yi ngay wax, du xaar nga jeexal sa frase
Jàppalekatu baat yi dañuy bind komand yi ci diir bu gàtt fekk yaa ngi wax
Ndaje ci jamono dëgg ak transkripsioŋ woote fu paccu resultaa yi wara feeñ saa yu nekk
Risk yi ak balustrade yi
Jëfandikoo baat ci anam wu jaarul yoon ak niru ak nit dafay gëna yokk sudee nanguwul.
Jaar-jaar mën na wàññeeku ci aksan yi, dialect yi wala barab yu bari xumbaay.
Audio synthetik mën nañu ko jaawale ak wax ju dëggu sudee amul etiket bu leer.
Roadmap ngir samp gi
Wutal ndigal bu leer ngir jàpp baat bi, klone ko ak jëfandikoowaat ko.
Saytu kalite ci kàddukat yu bari ak anam yu bari ci ginaaw.
Mandargal kañ la nit wara xoolaat wala nangu ay génne.
Etiketu audio synthetik te nga denc dokimaa ci fimu bawoo ngir mëna lim.
Weyal di banneexu
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
Gis bi ci topp
Tàqale RNN bu am ñaari yoon
Laaj yi ñuy faral di laaj
What is RNN-Transducer Models?
RNN-Transducer (RNN-T) ab architecture buy xàmmee kàddu la buy saafara ñakk kattan gu gëna mag ci CTC - ñàkka mëna modele dependence yi ci digganté token yiy génne. Dafay dooleel lu bari ci 'live' xàmmee kàddu yi nga jëfandikoo bis bu nekk ci aparey bi.
Ban yamaleg CTC la RNN-Transducer di wax ci?
RNN-T dafay yokk benn reso buy wax luy waaja am ci ay token yu njëkk, dindi xalaatu CTC ni génne bu nekk moom boppam ci audio bi.
Lan mooy ñetti mbir yi gëna am solo ci RNN-Transducer?
RNN-T dafay boole encodeur audio ak reso buy wax luy waaja am ci bind, boole ci reso bu ndaw buy joxe poñ yi ci topp.
Ban cër la reso biy wax luy am ci RNN-T?
Reseau biy wax luy waaja am dafay dox ni xeetu làkk bu biir ci kaw taarixu token biy génne, di jox RNN-T ortograafi ak motif baat yu dëggu.
Lan moo waral RNN-T di jàppale streaming bu woyof ci anam wu natureel?
RNN-T dafay dox ci réseau monotone time-by-token, suko defee mu mëna génne ay output yu audio bi yegsi ak latency bu yam moo gën ñu xaar clip bi yépp.
Ci dekodaasu RNN-T, lan mooy génne token bu amul dara?
Ci biir réseau RNN-T, ab blank dafay awaase ci axis waxtu bi (demal ci kadre audio bi ci topp), waaye ab non-blank dafay awaase ci axis etiket bi.