Ubuyobozi bwa Audio AI

ECAPA-TDNN Kumenyekanisha Umuvugizi

ECAPA-TDNN ni urusobe rw'imyubakire ihindura imiterere iyo ari yo yose ihindura clip iyo ari yo yose ikavamo imvugo yerekana amajwi, igafasha imashini kumenya uwuvuga.

2 min somaIbiherutse kuvugururwa

Incamake

It set the state of the art for speaker verification and remains the workhorse behind voice ID systems today.

Kwibira cyane

ECAPA-TDNN bisobanura Umuyoboro Wibanze Kwitondera, Kwamamaza no Guteranya mu gihe-Gutinda Neural Networks, yatangijwe na Desplanques na bagenzi be muri 2020. Yubakiye ku buryo bwa kera bwa x-vector ariko ikongeramo ibintu bitatu by'ingenzi byavuguruwe: Squeeze-Excitation ibuza guhuza imiyoboro ihanitse, ihuza ibice byinshi hamwe n’ibice byimbitse. Imvugo muburyo bumwe. Yatojwe hamwe ninyongera-margin softmax (AAM-softmax) igihombo kuri corpora nini nka VoxCeleb, itanga ibyapa aho clips imwe ya disikuru ihuriweho neza. Amajwi abiri agereranwa na cosine isa. Ku kizamini cya VoxCeleb1 cyashyizeho igipimo cyamakosa angana munsi ya 1 ku ijana, gusimbuka gukomeye kuri sisitemu zabanjirije iyi.

Ubushishozi

Amayeri yibanze ni imibare yitondewe guhuriza hamwe: aho kugirango ugereranye gusa urwego-urwego rwurwego, urusobe rwiga kuri buri murongo wibipimo byerekana uburemere bwibintu byingenzi (imvugo isobanutse neza) ibara kuruta guceceka cyangwa urusaku, hanyuma ikabara ikigereranyo kiremereye kandi kiremereye gutandukana. SE ihagarika hamwe na Res2Net-yuburyo bwinshi-buringaniza imyanzuro ireke buri cyiciro kimeze kumiterere yisi yose. Kwinjiza kwanyuma mubisanzwe ni ibipimo 192, byatsinzwe nintera ya cosine.

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 ECAPA-TDNN Kumenyekanisha Umuvugizi

Ubushakashatsi burimo kugana imbere-yiyobora-imbere nka WavLM na wav2vec 2.0 igaburira ECAPA-imiterere yinyuma-yinyuma, igabanya ibimenyetso byanditse bikenewe kandi ikongerera imbaraga urusaku na clips ngufi. Tegereza kwishyira hamwe hamwe na anti-spoofing kuburyo moderi imwe yerekana kandi ikemeza uwatanze disikuru, verisiyo ntoya itandukanijwe kugirango ikoreshwe ku bikoresho, hamwe n’ubutabera bukomeye kugira ngo ugabanye icyuho cy’amakosa ku nyito, imyaka, n'indimi uko biometrike y'ijwi yaguka muri banki no kugenzura uburyo.

Gushyira mu bikorwa Isi

Ijwi ryibinyabuzima byinjira muri banki ya terefone, aho ijwi ryumuhamagaye rihuye nicyitegererezo cyanditse aho kuba PIN.

Gutandukanya disikuru mugukoresha ibikoresho byo kwandikirana, kuranga 'uwavuze iyo' muguhuza ECAPA.

Forensic na call-center disikuru kugenzura kugirango ibendera niba amajwi abiri yaturutse kumuntu umwe.

Guha imbaraga disikuru-kugenzura ibisubizo mubikoresho bifunguye nka SpeechBrain na Kaldi kubashakashatsi no gutangira.

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 ECAPA-TDNN Speaker Recognition 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

Kumenyekanisha amajwi

Ibibazo bikunze kubazwa

What is ECAPA-TDNN Speaker Recognition?

ECAPA-TDNN ni urusobe rw'imyubakire ihindura imiterere iyo ari yo yose ihindura clip iyo ari yo yose ikavamo imvugo yerekana amajwi, igafasha imashini kumenya uwuvuga. Yashyizeho imiterere yubuhanzi bwo kugenzura abavuga kandi ikomeza kuba ifarashi ikora inyuma ya sisitemu yijwi rya none.

Nibihe bisohoka byibanze bya ECAPA-TDNN ya clip yatanzwe?

ECAPA-TDNN ishushanya imvugo ihindagurika-ndende mumurongo umwe uhamye-ushiramo icyerekezo cyerekana ijwi ryumuvugizi.

Nigute ibyapa bibiri bya ECAPA-TDNN mubisanzwe ugereranije no guhitamo niba ari abavuga rumwe?

Nyuma yo gushiramo, gukuramo cosine bisa (cyangwa amanota ajyanye nayo nka PLDA) bipima uburyo amajwi yombi yegeranye.

Niki 'kwitondera imibare ihuza' igikora?

Ibarurishamibare ryitondewe rihuza ibyiciro byitondewe byibanze kumurongo no kubiteranya muburyo buremereye kandi butandukanijwe, bushimangira kumurongo wamakuru.

Ni ubuhe butumwa bukoreshwa cyane mu guhugura no gupima urugero rwa ECAPA-TDNN?

VoxCeleb, igice kinini cyibiganiro byamamare byakuwe kuri videwo, ni corpus isanzwe yo guhugura no gusuzuma sisitemu yo kugenzura abavuga.

Niki 'SE' (Squeeze-Excitation) guhagarika bigira uruhare mubwubatsi?

Guhagarika-Kwishima biga kwiga gupima buri muyoboro wihariye ukoresheje amakuru yisi yose, kureka umuyoboro ushimangira imiyoboro yingirakamaro.