Ukwenziwa Kwezwi Lombhalo Kwenkulumo
Ukujwayela kombhalo isinyathelo esisekupheleni esiphinda sibhale umbhalo ongahluziwe ube amagama akhulunywa ngokuphelele ngaphambi kokuthi uhlelo lwenkulumo lusho.
Uhlolojikelele
It is what turns '$5' into 'five dollars' and '12/5/2024' into a spoken date, and getting it wrong is one of the most jarring TTS failures.
I-Deep Dive
Umbhalo obhaliwe ugcwele amagama angewona ajwayelekile: izinombolo, uhlobo lwemali, amadethi, izikhathi, izifinyezo, ama-URL, nezimpawu okungekho muntu oziphimisa njengoba zinjalo. Ukujwayela kombhalo (ngezinye izikhathi okubizwa ngokuthi i-TN front-end) kunweba lokhu kufomu lakho lezwi ukuze imodeli engezansi yomfula yazi ukuthi ithini ngempela — '$5' iba 'amadola amahlanu,' 'uDkt.' kuba 'udokotela' noma 'shayela' kuye ngomongo, futhi 'IV' ingase ibe 'ne,' 'i-intravenous,' noma izinhlamvu 'I-V.' Amasistimu endabuko asebenzisa imithetho ebhalwe ngesandla kanye nama-finite-state transducers (WFSTs), athembekile futhi afundekayo. Izindlela ezintsha zisebenzisa amamodeli we-neural sequence-to-sequence, kodwa i-neural TN emsulwa ingaveza amaphutha ayingozi (isho inombolo engalungile), ngakho-ke amasistimu okukhiqiza avame ukusebenzisa imiklamo eyingxube enemithetho njengezinsimbi zokuqapha. Ukuzwela kokuqukethwe kuyingxenye enzima: ithokheni efanayo ikhuluma ngokwehlukile kuye ngendawo ekuzungezile.
I-Technical Insight
Ukujwayela okujwayelekile kuqala kubeka amathokheni futhi kuhlukanise ithokheni ngayinye esigabeni se-semiotic (ukhadinali, idesimali, idethi, imali, isilinganiso, isifinyezo), bese kusetshenziswa isisho esiqondene nekilasi, esivame ukwakhiwa njengesiguquli esinesisindo esilinganiselwe esisheshayo futhi esihlolekayo ngokugcwele. Amathokheni angaqondakali awacaciswanga kusetshenziswa umongo wendawo kanye nengxenye yenkulumo yezimpawu. Amasistimu we-Neural kanye ne-hybrid ayibeka njengokubhala kabusha kombhalo kuya kombhalo kodwa okukhiphayo okuphoqayo - isibonelo, ukumboza uhlelo lolimi noma 'ukumaka bese kunwetshwa' - ukuvimbela amaphutha angamukeleki njengokufunda unyaka njengenombolo yocingo.
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 Lokujwayela Kombhalo Wenkulumo
Ukujwayela kuthrenda ku-neural-and-rule hybrids egcina ukuphepha kolimi lwesimo esilinganiselwe kuyilapho kusetshenziswa amamodeli afundiwe ukuze kuxazululwe umongo, kanye namamodeli olimi amakhulu aphatha okungcolile, umbhalo womhlaba wangempela nezilimi eziningi ngesikhathi esisodwa. Ucwaningo lugxile ekususeni amaphutha 'angenakulungiseka' kanye naku-TN yezilimi eziningi lapho amanani, idethi, nohlobo lwemali kuhluka kakhulu. Njengoba i-TTS esuka ekupheleni idonsa imisebenzi eminingi ephuma phambili, lindela ukujwayela kuhlale kuyisiteji esilawulekayo, esifundekayo ngokunembile ngoba amaphutha lapha abonakala kakhulu futhi abiza kakhulu.
Ukuqaliswa Komhlaba Wangempela
Ukufunda ngokuzwakalayo '$1,250.50' ngokuthi 'amadola ayinkulungwane namakhulu amabili namashumi amahlanu namasenti angamashumi amahlanu' kumsizi wezwi wasebhange.
Ukunweba izifinyezo ukuze 'St.' kukhulunywa ngokuthi 'umgwaqo' noma 'ungcwele' kuye ngomongo ekwazisweni kokuzulazula.
Ukukhuluma ngezinsuku, izikhathi, nezinombolo zocingo ngendlela efanele kuzinhlelo zokusebenza zekhalenda nezikhumbuzo.
Ukuguqula izimpawu namayunithi afana no-'5 km' noma '%' abe amagama akhulunywayo ezifundi zesikrini namathuluzi okufinyelela.
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 Text Normalization for Speech 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
Umbhalo oya Enkulumweni
Imibuzo evame ukubuzwa
What is Text Normalization for Speech?
Ukujwayela kombhalo isinyathelo esisekupheleni esiphinda sibhale umbhalo ongahluziwe ube amagama akhulunywa ngokuphelele ngaphambi kokuthi uhlelo lwenkulumo lusho. Yilokho okuguqula u-'$5' abe 'amadola amahlanu' kanye no-'12/5/2024' kube idethi enkulunywe ngayo, futhi ukuwenza kungalungile ukuhluleka kwe-TTS okuxaka kakhulu.
Kwenziwani ukujwayezwa kombhalo wenkulumo ngokuyinhloko?
Ukujwayela kunweba amathokheni angajwayelekile njengezinombolo, amadethi, nezifinyezo zibe ngendlela yazo eshiwo ngamazwi ngaphambi kokuhlanganiswa.
Ingabe uhlelo oluhle kufanele lujwayeze kanjani u-'$5' wenkulumo?
Imali idinga ukuhlela kabusha kanye nokusho amagama uphawu, ngakho-ke u-'$5' ukhulunywa 'njengamadola amahlanu,' hhayi ukusuka kwesokudla kuye kwesokudla.
Kungani kujwayelekile ithokheni efana 'noDkt.' noma 'IV' ekhohlisayo?
'UDkt.' kungaba 'udokotela' noma 'shayela' futhi 'IV' kungase kube 'okune,' 'i-intravenous,' noma izinhlamvu — umongo unquma ukuphinyiswa kwezwi okulungile.
Yibuphi ubuchwepheshe bendabuko obusetshenziswa kabanzi ukwakha imithetho ethembekile, efundekayo yokujwayelekile?
Ama-WFST afaka ama-grammars aklanywe ngesandla ashesha futhi ahloleka ngokugcwele, okuwenza abe yinketho yesikhathi eside yokukhiqiza i-TN.
Kungani amasistimu okukhiqiza evame ukugwema ukujwayela kombhalo we-neural?
I-Neural TN engaphoqelekile ingakhiqiza imiphumela engalungile kodwa eyingozi (isb., isibalo esingalungile), ngakho-ke imithetho isebenza njengonogada kumasistimu ayingxubevange.