GUIDE IA Audio

Moshi Diskur bu mat sëkk

Moshi source open-source la, IA baat ci jamono dëgg bu bawoo ci Kyutai, di wax ak déglu benn yoon - full-duplex - ludul jël turn yu tar.

2 simili jàngDañu mujjee yeesal

Résumé

That removes the awkward lag and rigid turn-taking of traditional voice assistants.

Plongeur bu xóot

Moshi, bi laboratwaaru Kyutai bu français genne ci 2024, xeetu fondasioŋ la buy wax-ci-kaddu, ñu tabax ko ngir waxtaan bu natureel te du yàgg. Moshi wuute na ak assistant pipeline yiy boole wax-ci-tekst, ginaaw ga modelu làkk, ginaaw ga bind-ci-kaddu, Moshi dafay jëfandikoo audio bi ci saasi te baña bàyyi. Xalaat bu mag bi mooy duplex bu mat: dafay modele ñaari stream audio ci benn yoon - jëfandikukat bi ak boppam - suko defee mu mëna déglu bi muy wax, jëflante ak dagg, backchannel ak 'mhm,' ak overlap ci nature ni nit ñi di defee. Dafay yegg ci diggante 160 ba 200 milisegond, di gëna suufeel lag assistant bi ñuy faral di def. Ci suufu kapot bi dafay boole ab modelu làkku bind ak audio bu am paramet 7B (Helium) ak Mimi, ab codec audio neural buy tënk kàddu yi ci ay token yu wuute yu model bi mëna defar. Kyutai dafa bàyyi poid yi ak kode bi ci kanam ñépp.

Gis-gis xarala

Kaf gi Moshi am mooy kodek Mimi bi, mooy soppi audio bu wéy ci bitrate bu woyof ci ay token yu wuute ci 12.5 Hz, boole ci benn token semantik buñ distiye. Royuwaayu lakk bi defay wax luy waaja am ci ay token yu jëfandikukat bi ci ay jamono yu tolloo, kon nak jamono ji du musa taxaw ngir 'déglu.' Xeetu 'Inner Monologue' dafay wax luy waaja am ci bind bi balaa audio bi, gëna baaxal kalite làkk wi ak lëkkaloo li Moshi wax dëgg.

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 Moshi Kaddu Duplex

Modeling full-duplex mingi nekk royukaay ci IA baat bu natureel, di am njeexital ci sistem yi ci liggéey bi yépp. Xaarandil versioŋ yu gëna ndaw, yu nekk ci aparey bi, jàppale ci làkk yu bari, latency bu néew, ak boole ci ndawu liggéey yi, serwiisu kiliyaan yi, ak jumtukaayi yombal jëfandikoo gi. Ndax Moshi ubbeeku na, gëstukat yi mën nañu ko saytu ak gëna suqali ko ci seen bopp. Jafe-jafe yi ñu ngi des ci wàllu wóolu dëgg, kaaraange ci wax yuy jaxasoo, ak nuance yu am yëg-yëg, waaye coppite bi joge ci jël turn-taking bu dëgër dem ci waxtaan bu yomb, buy dagg, mën na nekk luy yàgg.

Doxal ci àdduna dëgg

Baat bu amul loxo mën nga dagg ci diggu frase, ak tontu ci diir bu néew 200 milisegond.

Ubbi gëstub baseline ngir jàngat ci jamono dëgg, waxtaan bu mat sëkk te amul boyet yu ñuul yuñ moomal.

Jàppalekatu yombal jëfandikoo gi di waxtaan bu baax ak jëfandikukat yi soxla dellu ak dellu ci lu gaaw te baax.

Prototyping ay bot yuy wax ci serwiisu kiliyaan yi, yuy backchannel ak di react bi ki koy woo di wax ba leegi.

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 Moshi Full-Duplex Speech 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

Normalisasioŋu mbind ngir wax

Laaj yi ñuy faral di laaj

What is Moshi Full-Duplex Speech?

Moshi source open-source la, IA baat ci jamono dëgg bu bawoo ci Kyutai, di wax ak déglu benn yoon - full-duplex - ludul jël turn yu tar. Loolu dina dindi lag bu jaxasoo bi ak turn-taking bu dëgër bi assistant vocal yi daan am.

Luy 'full-duplex' ci wàllu Moshi?

Full-duplex dafay tekki ni model bi dafay jëfandikoo audio biy dugg ak biy genn benn yoon, suko defee mu mëna déglu bi muy wax te mëna jëfandikoo ay dagg ci boppam.

Ban mbootaay moo bàyyi Moshi?

Kyutai moo defar Moshi ba noppi ubbi ko, di labo gëstukat ci juntuwaay yu bees yu français, ci 2024.

Luy Mimi ci sistemu Moshi?

Mimi mooy codec audio neuronal biy dajale kàddu yu wéy yi ci biir benn bitrate bu woyof ci ay token yu wuute yu model bi mëna defar.

Lan moo wuutale Moshi ak gasoduk assistant vocal bu yàgg bi?

Jàppalekatu cosaan yi dañuy boole wax-ci-tekst, xeetu làkk, ak mbind-ci-kaddu; Moshi benn xeetu wax-ci-kaddu la buy jëfandikoo audio bi saa yu nekk.

Lu tollu ci ban latency ci waxtaan la Moshi di wut?

Moshi mingi am latency buy tollu ci diggante 160 ba 200 ms, di gëna suufeel assistant vocal yi ñuy faral di def, loolu mooy tax ñu mëna jaxasoo ak dagg.