Mimi Streaming Audio Codec
Mimi ni codec yamajwi yerekana amajwi agabanya imvugo mumurongo muto wibimenyetso bya discret mugihe nyacyo, kubwibyo moderi ya AI irashobora kumva no kuvuga nubukererwe buke cyane.
Incamake
It is the audio backbone behind Kyutai's Moshi voice model.
Kwibira cyane
Mimi, yasohowe na laboratoire y’Abafaransa Kyutai mu 2024, ni codec ya neural ihindura amajwi 24 kHz mu muyoboro w’ibimenyetso bya disiketi hafi ya kbps 1.1 na 12.5 gusa ku isegonda. Ikoresha kodegisi-decoder hamwe na vector isigaye (RVQ), igabanya ibimenyetso murwego rwa 'semantique' urwego rwa mbere rutandukanijwe nicyitegererezo cyo kuvuga (WavLM) hiyongereyeho urwego rwinshi rwa 'acoustic' rufata amajwi. Byibanze byuzuye kandi bitera: bisohora ibimenyetso nkuko amajwi ageze aho gutegereza clip yuzuye, hamwe na ms 80 z'ubukererwe. Ibi bituma ururimi rwicyitegererezo rufata imvugo nkibimenyetso byanditse, bigafasha Moshi kuganira muri duplex yuzuye mugihe amajwi yubatswe yubatswe yumvikana kandi asanzwe.
Ubushishozi
Amayeri ya Mimi ni gahunda yo gutandukana-RVQ. Igitabo cya mbere codebook yatojwe hamwe no gutakaza igihombo kugirango ihuze ibyashizwe muri WavLM, bikayihatira gutwara 'ibisobanuro' bya fonetike, mugihe ibitabo bisa na acoustic codebooks byubaka ibisobanuro birambuye. Transformer ikorera imbere mu cyuho, kandi igihombo cyo guhangana (GAN) kuri decoder gikarisha ubwiza bwibisohoka. Impamvu zitera zituma ibintu byose bigenda neza, bityo ubukererwe buguma hafi ms 80.
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 Mimi Streaming Audio Codec
Tegereza codecs nka Mimi guhinduka intera isanzwe hagati yijwi nururimi runini, usunikire abafasha kumajwi nyayo mugihe cyo gusubiza munsi ya 100 ms. Ubushakashatsi butwara ibipimo byerekana ibimenyetso biri hasi mugihe urinda indangamuntu, amarangamutima, numuziki. Kuberako Kyutai ifunguye-Mimi na Moshi, birashoboka ko izatera imbuto nyinshi zifungura imvugo-zivuga, abafasha ku bikoresho, hamwe nibikoresho byitumanaho byijwi rya ultra-low-bandwidth.
Gushyira mu bikorwa Isi
Guha imbaraga Moshi ya Kyutai yuzuye-duplex umufasha wijwi kugirango ishobore kumva no kuvuga icyarimwe
Gutondeka imvugo yerekana urugero rwururimi mugihe nyacyo cyo kuvuga-gusobanura
Ultra-hasi-bitrate ijwi rihamagara (~ 1.1 kbps) kubintu bikennye cyangwa byuzuye urusobe
Gufata amajwi kubijyanye no kubyara imvugo hamwe nimiyoboro-y-imvugo itanga ibitekerezo hejuru yijwi nkinyandiko
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
Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.
Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.
Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.
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 Mimi Streaming Audio Codec quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ubuyobozi bukurikira
Kode ya Audio
Ibibazo bikunze kubazwa
What is Mimi Streaming Audio Codec?
Mimi ni codec yamajwi yerekana amajwi agabanya imvugo mumurongo muto wibimenyetso bya discret mugihe nyacyo, kubwibyo moderi ya AI irashobora kumva no kuvuga nubukererwe buke cyane. Numugongo wamajwi inyuma ya Moshi yijwi rya Kyutai.
Ni izihe laboratoire yarekuye Mimi na moderi y'ijwi rya Moshi?
Mimi na Moshi bararekuwe mu 2024 na laboratoire y’ubushakashatsi y’Abafaransa Kyutai, ifungura byombi.
Hafi ya bangahe ku isegonda Mimi asohora kuvuga?
Mimi ikanda amajwi 24 kHz kugeza kumurongo wa 12.5 gusa kumasegonda hafi 1.1 kbps.
Niki igitabo cya mbere ('semantique') codebook muri Mimi gifata?
Urwego rwa mbere rwa RVQ rwahuguwe hakoreshejwe distillation kuva WavLM kugirango itware ibintu bisobanutse / fonetike, mugihe urwego rwanyuma rwongeramo ibisobanuro bya acoustic.
Kuki Mimi adondorwa nk '' gutembera 'cyangwa' impamvu '?
Mimi itunganya amajwi kubwimpamvu kandi isohora ibimenyetso byiyongera, itanga ubukererwe bwa ms 80 bukwiranye no kuganira neza.
Nubuhe buryo bwo kubara Mimi akoresha mugushiraho amajwi?
Mimi ikoresha igereranya rya vector isigaye, igashyiraho codebook nyinshi kuburyo buriwese yongeraho ibisobanuro byiza kubibanjirije.