Ubuyobozi bwa Audio AI

Guhuza by'agateganyo Ibyiciro

Connectionist Temporal Classification (CTC) nuburyo bwo gutakaza nuburyo bwa decoding butuma imiyoboro yimitsi ihindura amajwi maremare mumajwi ntamuntu uhuza amajwi kuri buri nyuguti.

2 min somaIbiherutse kuvugururwa

Incamake

It made end-to-end speech recognition practical by solving the brutal alignment problem.

Kwibira cyane

Imvugo ni akajagari: ijambo 'uraho' rishobora kumara amajwi 40 y'amajwi, kandi ntamuntu numwe wanditse neza ikadiri ari 'h'. CTC, yatangijwe na Alex Graves mu 2006, ikuraho ibi. Umuyoboro usohora ibishoboka hejuru yinyuguti (hiyongereyeho ikimenyetso cyihariye 'ubusa' kuri buri kintu. CTC noneho isobanura guhuza byemewe nkikintu icyo aricyo cyose kumurongo-kumurongo ugwa kumurongo wanditse nyuma yamategeko abiri: guhuza inyuguti zasubiwemo, hanyuma ugasiba ubusa. Kuberako inzira nyinshi zishushanya kumyandiko imwe, CTC yerekana amahirwe ya bose ukoresheje algorithm ya dinamike-progaramu ya algorithm (imbere-inyuma-algorithm) kandi igatoza umuyoboro kugirango ugere kuri byinshi. Ikimenyetso cyambaye ubusa ni amayeri yubwenge areka icyitegererezo kivuga ngo 'nta gishya hano' kandi gitandukanya gusubiramo kwukuri nka kabiri-L muri 'muraho'.

Ubushishozi

Intego nyamukuru ya CTC ni ubwigenge buteganijwe: ukurikije amajwi, ibisohoka buri kadamu byahanuwe byigenga, nta moderi yindimi yatetse. Ibyo bituma incamake yimbere-isubira inyuma ikurikirana ariko bivuze ko CTC ikunda gutanga umusaruro ushimishije, cyane cyane (ufite ubusa, ufite inyuguti zikarishye) hamwe ninyungu zituruka kumurongo wururimi rwo hanze mugihe cya decode. Gushakisha ibiti hamwe na LM yahujwe, bakunze kwita prefix-beam decoding, bitezimbere kuburyo bugaragara kurenza umururumba wa argmax.

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.

Ejo hazaza h'umuhuza w'agateganyo

CTC ikomeje kuba ifarashi ikoreramo, cyane cyane aho ibintu bitemba kandi bitinda, kandi bigenda bikoreshwa nkigihombo cyingirakamaro hamwe no kwitabwaho cyangwa intego za transducer muburyo bwa Hybrid 'CTC / kwitondera'. Tegereza CTC gukomeza nk'ishami ryihuta, ryoroshye rya decoder imbere muri sisitemu nini yo kuvuga, kandi nka moteri yo guhuza inyuma yibikoresho-bihuza ibikoresho byerekana igihe. Kiyobora wenyine-kodegisi nka wav2vec 2.0 mubisanzwe bihujwe neza numutwe wa CTC.

Gushyira mu bikorwa Isi

Guhuza neza wav2vec 2.0 hamwe numutwe wa CTC kugirango wubake isoko-ifunguye imvugo-y-inyandiko-mvugo mu rurimi ruke

Kubyara ijambo- na foneme-urwego rwibihe bya subtitles na karaoke ukoresheje CTC ku gahato

Igihe nyacyo cyanditse ku gikoresho aho moderi ya CTC itembera hamwe nubukererwe buke

Kumenyekanisha intoki, aho CTC isoma umurongo wo gutukana utabanje gutandukanya inyuguti zitandukanye

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 Connectionist Temporal Classification 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

Ibyiciro byumuziki

Ibibazo bikunze kubazwa

What is Connectionist Temporal Classification?

Connectionist Temporal Classification (CTC) nuburyo bwo gutakaza nuburyo bwa decoding butuma imiyoboro yimitsi ihindura amajwi maremare mumajwi ntamuntu uhuza amajwi kuri buri nyuguti. Byatumye imvugo irangira-iherezo kumenyekanisha bifatika mugukemura ikibazo cyubugome bukabije.

Ni ikihe kibazo cy'ibanze CTC yagenewe gukemura kugirango imenye imvugo?

CTC ireka umuyoboro wa gari ya moshi kuri (amajwi, inyandiko) kubiri ntawe uranga ikadiri ihuye nimiterere, ikemura ikibazo cyo guhuza.

Niki kimenyetso kidasanzwe 'ubusa' muri CTC gikoreshwa?

Ikimenyetso cyambaye ubusa kireka icyitegererezo gisohora 'ntakintu gishya' kumurongo kandi, cyane, gitandukanya gusubiramo kwukuri nka kabiri-L muri 'uraho' nibisubirwamo.

Nigute CTC ibara igihombo cyo kwandukura intego?

CTC ikoresha imbere-isubira inyuma dinamike-progaramu ya algorithm kugirango igereranye ibishoboka kuri buri murongo ugereranya ikarita nyuma yo guhuza gusubiramo no gusiba ubusa.

Ni ayahe mategeko abiri asenyuka CTC ikoresha kugirango ihindure inzira-inzira mumyandiko yanyuma?

Inzira mbisi kuri buri kintu cyacishijwe bugufi no kubanza guhuza inyuguti zibiri zegeranye hanyuma zigakuraho ibimenyetso byose byubusa.

Niki cyoroshya gutekereza CTC isanzwe ikora kubyerekeye ibisubizo byayo?

CTC ifata ubwigenge bushingiye ku bwigenge, butuma incamake ikurikirana ariko bivuze ko nta cyitegererezo cyururimi rwubatswe.