Ubuyobozi bwa Audio AI

Wav2Inyandiko Yemeza ASR

Wav2Letter ni sisitemu yo kumenyekanisha imvugo iherezo-iherezo kuva kuri Facebook AI yakoresheje imiyoboro ihamye gusa, ntisubire.

2 min somaIbiherutse kuvugururwa

Incamake

It mattered as a fast, simple alternative that proved CNNs alone could transcribe speech competitively.

Kwibira cyane

Wav2Letter yatangijwe nubushakashatsi bwa Facebook AI mu 2016, yavuye muburyo bwiganje kandi bushingiye kuri HMM yishingikiriza rwose kumiyoboro yimitsi ihuza amakarita yerekana amajwi inyuguti (inyuguti), bityo izina. Ubusanzwe yatojwe hamwe na AutoSegCriterion yihariye (ASG) igihombo, ubundi buryo bworoshye kubihombo bisanzwe CTC yataye ikimenyetso cyubusa hamwe nicyitegererezo cyinzandiko. Byanditswe muri C ++ ukoresheje Flashlight / ArrayFire inyuma, byakozwe muburyo bwihuse kuri CPU na GPU. Inyandiko zanyuma, Wav2Letter ++ hamwe nuburyo bwuzuye bwo kwemeza, byapimye kuri datasets nini kandi bigera ku gipimo cyamakosa yamagambo kuri Librispeech. Igishushanyo cyacyo gusa cyashushanyaga cyane kandi kigereranywa na decodeur ya RNN ikurikiranye.

Ubushishozi

Wav2Letter ikusanya 1D ihindagurika ryigihe gito hejuru yimiterere ya acoustic, hamwe na buri cyiciro cyagura umurima wakira kuburyo ibice byimbitse bifata imiterere ndende ntisubire. Kuberako imyanzuro itunganya igihe cyose intambwe ibangikanye, imyitozo no gufata umwanzuro byihuse. Igihombo cyumwimerere ASG isa na CTC ariko ikuraho ikimenyetso cyambaye ubusa kandi ikongeramo amanota yinzandiko-yinzibacyuho yerekana amanota, ikabyara ibipimo bitandukanye bitandukanye bikurikirana bihuza amajwi-ahinduranya amajwi n'amajwi asohoka nta kirango kirimo.

Ingaruka z'Ingamba

Kugera no kugera

Itezimbere kugerwaho binyuze mu kwandukura, kuvuga, no guhuza amajwi.

Igiciro na bije

Amatsinda yibitangazamakuru arashobora kohereza amajwi yihuse hamwe na bije nto.

Umuvuduko n'igipimo

Sisitemu ireba abakiriya irashobora gutunganya imikoranire ivugwa murwego runini.

Kazoza ka Wav2Inyandiko Yemeza ASR

Imirongo itaziguye ya Wav2Letter iba muri Flashlight, isomero ryiga imashini ya C ++ ya Facebook, kandi imenyesha moderi ya wav2vec yiyobora ubu yiganje. Isomo ryagutse, ibyo kwemeza hamwe nububiko bubangikanye birashobora guhura nibisubirwamo, bigaburirwa muburyo bwa ASR. Tegereza sisitemu zizaza kugirango ukomeze kuguza Wav2Urwibutso rwibanda kumiyoboro ikora neza, iringaniye, itandukanye rwose itangirira kumpera-iherezo mugihe ushyira mugikorwa cyo kwiyobora kwitegura indimi nkeya.

Gushyira mu bikorwa Isi

Igihe-nyacyo cyo kwandukura aho gutinda-gutinda, kugereranya kugereranya bifite agaciro kuruta ingingo nke zukuri

Kuri-igikoresho cyangwa CPU imenyekanisha imvugo idashobora kugura decoder ziremereye

Ibyingenzi byubushakashatsi ugereranya ASR yemewe na RNN hamwe na sisitemu ya transformateur kuri Librispeech

Gukora nkibishingiro byububiko bwibitabo bya Flashlight ya Facebook hanyuma moderi ya wav2vec

Ingaruka & Kurinda

Gukoresha nabi amajwi no kwigira ibyago byiyongera mugihe uruhushya rubuze.

Ukuri kurashobora kugabanuka hejuru yimvugo, imvugo, cyangwa urusaku rwibidukikije.

Amajwi yubukorikori arashobora kwibeshya kumvugo yukuri nta kirango gisobanutse.

Igishushanyo mbonera

1

Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.

2

Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.

3

Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.

4

Andika amajwi yubukorikori kandi ugumane inyandiko zerekana kubazwa.

Komeza Ubushakashatsi

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 Wav2Letter Convolutional ASR quiz

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

Tangira ikibazo

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

Ubuyobozi bukurikira

Imiyoboro ya Neural Network

Ibibazo bikunze kubazwa

What is Wav2Letter Convolutional ASR?

Wav2Letter ni sisitemu yo kumenyekanisha imvugo iherezo-iherezo kuva kuri Facebook AI yakoresheje imiyoboro ihamye gusa, ntisubire. Byari bifite akamaro nkuburyo bwihuse, bworoshye bwerekanaga CNNs yonyine ishobora kwandukura imvugo kurushanwa.

Ni ubuhe bwoko bw'imyubakire ya Wav2Letter yishingikiriza gusa?

Wav2Urwandiko rurazwi mugukoresha gusa imiyoboro yimitsi ihindagurika, irinda kugaruka burundu.

Ni irihe shyirahamwe ryateje imbere Wav2Letter?

Wav2Letter yatangijwe nubushakashatsi bwa Facebook AI muri 2016 nyuma iza guhinduka isomero rya Flashlight.

'Inyuguti' muri Wav2Urwandiko ruvuga iki?

Wav2Urutonde ushushanya amajwi yerekana amajwi muburyo bukurikiranye bwinyuguti (inyuguti), inzira yanyuma-iherezo.

Nigute igihombo cyambere AutoSegCriterion (ASG) gitandukaniye nigihombo gikunze kugaragara CTC?

ASG ita CTC yerekana ubusa hanyuma ikongeramo amanota yinzibacyuho hagati yinyuguti mugihe zisigaye zitandukanye.

Ni izihe nyungu zingenzi zifatika za Wav2Letter yogushushanya gusa?

Bitandukanye na RNN ikurikiranye, imyanzuro irashobora kubarwa mugihe kimwe, bigatuma sisitemu yihuta kandi ikora neza.