UMHLAHLANDLELA WE-AI womsindo

Lalela Bheka futhi Upela

Listen, Atttend and Spell (LAS) iyingqophamlando yenethiwekhi ye-neural yango-2015 ebhala inkulumo ngokuqondile ibe yizinhlamvu, ngaphandle kwesichazamazwi esakhelwe ngesandla sokuphimisa noma imodeli yolimi ehlukene.

2 amaminithi ukufundaIgcine ukubuyekezwa

Uhlolojikelele

It showed that a single end-to-end model could do speech recognition.

I-Deep Dive

I-Lalela, Yamela futhi Upele, yethulwe Google abacwaningi u-Chan, Jaitly, Le, kanye noVinyals ngo-2015, bekungesinye seziqapheli zeqiniso zokuqala zokuphela-ekugcineni. Inezingxenye ezimbili: 'Umlaleli,' i-LSTM eyi-pyramidal bidirectional LSTM ehlanganisa umsindo ngenkathi inciphisa ubukhulu besikhathi, kanye 'Nesipelingi,' isiqophi se-LSTM esisekelwe ekunakekelweni esikhipha izinhlamvu ngesikhathi esisodwa. Indlela yokunaka ivumela Isipelingi ukuthi sigxile ocezwini olufanele lomsindo wohlamvu ngalunye oluphumayo. Ngokungafani namapayipi amadala e-HMM-DNN, i-LAS ayidingi isichazamazwi sefonimu, akukho ukuqondanisa okuphoqelelwe, futhi ayikho imodeli yolimi eqeqeshwe ngokuhlukile; ifunda isipelingi, imingcele yamagama, nama-acoustics ngokuhlanganyela emsindweni olotshiwe. Ikhuthaze ngokuqondile izinhlelo ze-ASR zesimanje zokulandelana-kuya-kulandelana.

I-Technical Insight

I-LAS ihlanganisa i-encoder-decoder nokunaka. Isishumeki se-pyramidal LSTM sihlukanisa ukulinganisa kwesikhathi kungqimba ngayinye kwezintathu, sisika ukulandelana okude kwe-acoustic kube ubude obulawulekayo ukuze ukunaka kubambezeleke. Kuso sonke isinyathelo sokuqopha Isipelitha sibala izisindo zokunaka kuzo zonke izifundazwe zokufaka ikhodi, sizihlanganise zibe ivekhtha yomongo, futhi sibikezele uhlamvu olulandelayo. Ukuqeqeshwa kukhulisa amathuba okulandelana kwezinhlamvu ezifanele; iqhinga lesampula elihleliwe linciphisa ukungafani kwesitimela/ukuhlola.

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 Lalela Yiya Futhi Upelwe

I-LAS manje isiwumlando, kodwa i-DNA yayo isebenza kuzo zonke izinhlelo zesimanje ze-ASR. Umbono wayo osuselwe ekunakekelweni we-encoder-decoder uguquke waba yiziboni ze-Transformer ne-Conformer, kuyilapho izindlela ezihlobene ezifana ne-RNN-Transducer power on-device dictation. Amasistimu esikhathi esizayo aqhubekisela phambili lo mzila osuka ekupheleni ukuya ekupheleni, ehlanganisa ukuqashelwa nokuhumusha nokuqonda ngamamodeli ezilimi eziningi, futhi ephushela ekusakazeni, ukuloba okubambezeleke okuphansi i-LAS, ngenxa yokungasakazwa, engakwazi ukukunikeza ekuqaleni.

Ukuqaliswa Komhlaba Wangempela

Ukuloba isiNgisi esikhulunywayo ngqo ezinhlamvwini ngaphandle kwesichazamazwi sokuphimisa

Isebenza njengesisekelo somqondo sokubizelwa kwezwi okusekelwe ekunakekelweni kanye nezinhlelo zamagama-ncazo

Ukubonisa ukuqeqeshwa kokugcina kuze kube sekupheleni kwezifundo zokuqaphela inkulumo kanye namabhentshimakhi

Amamodeli akhuthazayo okulandelana-kuya-kulandelana asetshenziswa kamuva kumapayipi okuhumusha inkulumo

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

1

Thola imvume esobala yokuthwebula izwi, ukuhlanganisa, nokusebenzisa kabusha.

2

Ikhwalithi yokuhlola kuzo zonke izipikha nezimo zangemuva.

3

Chaza ukuthi kunini lapho umuntu kufanele abuyekeze noma agunyaze okuphumayo.

4

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 Listen Attend and Spell quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Qala imibuzo

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Umhlahlandlela olandelayo

Ukumaka Okuzenzakalelayo Komculo

Imibuzo evame ukubuzwa

What is Listen Attend and Spell?

Listen, Atttend and Spell (LAS) iyingqophamlando yenethiwekhi ye-neural yango-2015 ebhala inkulumo ngokuqondile ibe yizinhlamvu, ngaphandle kwesichazamazwi esakhelwe ngesandla sokuphimisa noma imodeli yolimi ehlukene. Kubonise ukuthi imodeli eyodwa ephuma ekupheleni ingakwazi ukuqaphela inkulumo.

Yiziphi izingxenye ezimbili eziyinhloko zemodeli ye-LAS?

I-LAS iqukethe isifaki khodi 'Somlaleli' esicubungula umsindo kanye nesikhiphi sekhodi esisuselwe ekunakekelweni 'Sokupela' esikhipha izinhlamvu.

Yini i-Speller ekuphumeni kwe-LAS?

I-Speller iyidekhoda ekhiqiza uhlamvu olulotshiwe ngohlamvu, ifunda isipelingi ngokuqondile.

Kungani isifaki khodi se-LAS sibizwa nge-'pyramidal'?

Isendlalelo ngasinye se-pyramidal BLSTM sinciphisa ubude bokulandelana ngohhafu, sifinyeze ukulandelana komsindo okude ukuze ukunaka kube nokwenzeka ngekhompyutha.

Iyiphi ingxenye endala i-LAS engayidingi ngokuphawulekayo?

Ngokungafani namaphayiphi e-HMM-DNN yakudala, i-LAS ifunda ngokuqondile kusukela kumapheya omsindo ukuya kumbhalo ngaphandle kwesichazamazwi sefonimu noma ukuqondanisa okuphoqelelwe.

Iyiphi indlela evumela ukuthi Isipelingi sigxile engxenyeni efanele yomsindo wohlamvu ngalunye?

Indlela yokunaka ibala izisindo ngaphezu kwezimo zesishumeki sekhodi, yakhe ivektha yomongo esetshenziswa isikhiphi sekhodi esinyathelweni ngasinye.