GUIDE IA Audio

Architecture konformer

Conformer bi ab bloku reso neuronal la buy boole convolution ak bàyyi xel ci sa bopp, di jàpp motif son yu baax ak contexte yu sori ci benn couche.

2 simili jàngDañu mujjee yeesal

Résumé

It became the de facto standard encoder for state-of-the-art speech recognition.

Plongeur bu xóot

Google moo ko dugal ci 2020, Conformer tontu na benn jafe-jafe bu am solo ci modeling audio: bàyyi xel ci sa bopp (ci Transformers) lu baax la ci contexte globaal waaye dafa néew doole ci motif yu am pepp yu sew yuy wuutale fonem yi, waaye convolution yi dañu gëna am doole ci barab bi. Blok Conformer bi daf leen di boole ci jëmmal 'sandwich': benn modulu feed-forward bu am genn-wàll, ginaaw ga benn modulu bopp yu bari, ginaaw ga benn modulu convolution, ginaaw ga ñaareelu modulu feed-forward bu am genn-wàll, ak normalisation couche ak lëkkaloo yu des ci biir. Module konvolusioŋ dafay jëfandikoo konvolusioŋ yuñ mëna tàqale ci xóotaayu ak benn yunit ligneer buñ tëj. Suñu boolee liggéey bu dëkk bi ak àdduna bi ci biir blok bu nekk, enkodeer Conformer yi dañuy wàññi njuumti yi ci baat yi ci kaw Transformer bu sell bi wala ligne yu sell yi ci kaw benchmark yu melni LibriSpeech.

Gis-gis xarala

Signature 'Macaron' bi dafay wër bàyyi xel bi ak convolution bi ci digganté ñaari couche feed-forward, bu nekk di joxe genn-wàll ci residu bi (facteur 0.5 bi), inspiré ci analyse Transformer FFN pairs. Module convolution bi dafay boole ab convolution pointwise ak ab aktivasioŋ GLU, ab convolution bu xóot, ab normalisasioŋ ci lots, ab aktivasioŋ Swish, ak ab convolution pointwise bu mujj - anam wu baax ngir modele contexte lokal te du jël lim parametre.

njeextalu pexe

Dugg ak yegg

Dafay gëna yombal jëfandikoo gi jaaraleko ci transkripsioŋ, nettali ak interfaasu baat.

Njëgg ak budget

Ekipu mejaa yi mën nañu yónnee audio bu leer ci anam wu gëna gaaw te seen xaalis gëna néew.

Gaawaay ak yaatuwaay

Sistem yiy jàkkarloo ak kiliyaan bi mën nañu def waxtaan ci anam wu gëna yaatu.

Ëlëgu architecture conformer

Leegi conformers yi ñooy liggéey ci encodeur yaxu ndigg ngir transducer ak CTC/attention ASR, te jëmmal tasaaroo ci tekki kàddu, xàmmee waxkat, ak gis xew-xew audio. Gëstu bu am solo dafay yombal bàyyi xel ci audio yu gudd (ligneer ak bàyyi xel ci streaming), distil Conformers ngir jëfandikoo ci aparey, ba noppi boole leen ak tàggat yaram bu ñuy saytu seen bopp. Xeet yu melni Squeezeformer ak Conformer bu baax dañuy gëna yokk njaayum njub ak xayma.

Doxal ci àdduna dëgg

Liggéey ni enkodeer ci defar sistem ASR ci ginaaw assistant yuy wax ak dikte

Xeetu tekki làkk yu am doole yuy tekki làkk wi ñuy wax, ba noppi tekki ko ci njeexte ba ci njeexte

Backbone ngir saytu waxkat bi ak bind ci surnaal, ràññee ki wax ci ndaje

Xew-xew audio ak xaaj son, lu ci melni gis alarm, wax, wala music ci biir benn stream

Risk yi ak balustrade yi

Jëfandikoo baat ci anam wu jaarul yoon ak niru ak nit dafay gëna yokk sudee nanguwul.

Jaar-jaar mën na wàññeeku ci aksan yi, dialect yi wala barab yu bari xumbaay.

Audio synthetik mën nañu ko jaawale ak wax ju dëggu sudee amul etiket bu leer.

Roadmap ngir samp gi

1

Wutal ndigal bu leer ngir jàpp baat bi, klone ko ak jëfandikoowaat ko.

2

Saytu kalite ci kàddukat yu bari ak anam yu bari ci ginaaw.

3

Mandargal kañ la nit wara xoolaat wala nangu ay génne.

4

Etiketu audio synthetik te nga denc dokimaa ci fimu bawoo ngir mëna lim.

Weyal di banneexu

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 Conformer Architecture quiz

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

Tambalil quiz

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

Gis bi ci topp

Architecture de parole

Laaj yi ñuy faral di laaj

What is Conformer Architecture?

Conformer bi ab bloku reso neuronal la buy boole convolution ak bàyyi xel ci sa bopp, di jàpp motif son yu baax ak contexte yu sori ci benn couche. Nekk na de facto encoder biñ gëna jëfandikoo ngir xàmmee kàddu yi.

Yan ñaari mekaniism la architecture Conformer di boole ci benn blok?

Conformer dafay boole convolution (ngir motif local) ak bàyyi xel ci sa bopp (ngir contexte global) ci biir bloc bu nekk.

Lan moo waral boole convolution ak attention lu am solo la ci wax?

Convolutions yi dañu gëna am solo ci siñaal yu ndaw yiy wuutale fonem yi, fekk bàyyi xel ci sa bopp dafay wane dependence ci wax ji yépp; Konformer amna ñoom ñaar.

Lan moy 'Macaron' wala jumtukaayu sandwich bu benn bloku Conformer?

Conformer dafay def module yiy bàyyi xel ci seen bopp ak yiy convolution ci diggante ñaari module feed-forward, bu nekk ci ñoom def ci residu bu am genn-wàll pondéré.

Ban mbootaay moo dugal Conformer, ak ci ban at?

Gëstukat yi Google ñoo dugal Conformer ci 2020 ci diir bu gàtt mu nekk enkodeer buy xàmmee kàddu yiñ miin.

Lan mooy modulu convolution bi ci biir Conformer bi?

Module konvolusioŋ dafay boole convolusioŋ pointwise ak benn unité ligneaire buñ tëj, convolusioŋ xóotaayu, normalisasioŋ lots, ak aktivasioŋ Swish, mujjee ci beneen conv pointwise.