VALL-E na Codec Ururimi rwicyitegererezo
VALL-E yahinduye inyandiko-y-ijambo nkikibazo cyo kwerekana ururimi hejuru yikimenyetso cyamajwi ya codec, ituma ijwi ryakoronijwe kuva amasegonda atatu gusa yicyitegererezo.
Incamake
It showed that the same next-token prediction powering text LLMs can generate remarkably natural, expressive speech.
Kwibira cyane
Byatangajwe na Microsoft mu ntangiriro za 2023, VALL-E ifata synthesis imvugo nko kwerekana imvugo. Aho guhanura spekrogramu, irahanura ibimenyetso bya acoustic discrets ya codec ya neural codec (EnCodec), bityo ibisekuru bihinduka ubutaha-ibimenyetso byerekana amajwi. Urebye amasegonda 3-yerekana amajwi atagaragara wongeyeho inyandiko yanditse, VALL-E ikomeza mwijwi ryuwo muvugizi, irinda timbre ndetse n’ibidukikije bya acoustic. Yahuguwe kumasaha agera ku 60.000 yo kuvuga, cyane cyane kuruta imibare isanzwe ya TTS, itanga clone ikomeye ya zeru. Kuberako ibimenyetso bya codec byashyizwe kumurongo (binyuze kuri RVQ), VALL-E ikoresha ibyiciro bibiri: moderi ya autoregressive ihanura icyambere, icyerekezo cyoroshye cyateganijwe kuri progaramu, kandi moderi itari autoregressive yuzuza ibimenyetso bisigaye. Iyi codec-LM resept yahumekeye abasimbuye nka VALL-E 2 hamwe nuburyo bwinshi bwo kuvuga.
Ubushishozi
Amayeri ni Hybrid decoding hejuru ya kodegisi ya codec. Icyiciro cya autoregressive gihanura ibyingenzi byambere-codebook ikimenyetso kimwekimwe, gufata prosody nibirimo. Ibitabo bisigaye byongeweho, byongeweho neza acoustic, byahanuwe muburyo bubangikanye nicyitegererezo kitari autoregressive cyateganijwe kumugezi wa mbere hamwe na disikuru yihuta. Uku gutandukana kugumana ubuziranenge mugihe wirinze ikiguzi cyo kubyara buri kimenyetso gikurikiranye, kandi gukoresha codec bisobanura imvugo ninyandiko bishobora kugereranywa nimashini imwe ya transformateur.
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 VALL-E na Codec Ururimi rwicyitegererezo
Ururimi rwimyandikire ya Codec ni uguhuza imvugo nicyitegererezo kinini cyururimi, yerekana sisitemu ihuriweho yumva, ibitekerezo, kandi ivuga muburyo bumwe. Witege ituze ryiza nibindi bihangano bike, ibihe nyabyo bigenda neza, hamwe no kugenzura neza amarangamutima nuburyo. Ihinduka rikomeye rituma VALL-E igira akamaro mugushikira no gushidikanya nayo itera impungenge zimbitse no kwemererwa, bityo gushyiramo amazi, kurinda amajwi-kugenzura, hamwe no kurinda politiki bigenda biba igice cyingenzi cyuburyo izo sisitemu zikoreshwa.
Gushyira mu bikorwa Isi
Guteranya ijwi kuva kumasegonda make yamajwi kubafasha kugiti cyabo cyangwa ibikoresho byabugenewe bigarura ijwi ryatakaye
Guhindura no gushiramo amashusho mu zindi ndimi mugihe ugumya kuvuga timbre yumwimerere
Gukora ibisobanuro byerekana, bihuye nibisobanuro bibika amajwi ya acoustic ibidukikije
Gukora nkumugongo wimvugo mubufasha ba multimodal bombi bumva kandi batanga amajwi yavuzwe
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 VALL-E and Codec Language Models 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
Ubushobozi bwihutirwa bwururimi runini
Ibibazo bikunze kubazwa
What is VALL-E and Codec Language Models?
VALL-E yahinduye inyandiko-y-ijambo nkikibazo cyo kwerekana ururimi hejuru yikimenyetso cyamajwi ya codec, ituma ijwi ryakoronijwe kuva amasegonda atatu gusa yicyitegererezo. Yerekanye ko kimwe-gikurikira-ibimenyetso byerekana imbaraga LLMs ishobora kubyara imvugo idasanzwe, yerekana.
Ni ikihe gitekerezo cyibanze gitandukanya VALL-E na sisitemu yambere-y-imvugo?
VALL-E ivugurura TTS nkibikurikira-byerekana ibimenyetso byerekana amajwi ya codec yerekana amajwi, bifata ibisekuruza byururimi nkicyitegererezo cyururimi.
Ni amajwi angahe yerekana VALL-E ikeneye guhuza ijwi ryumuvugizi mushya?
VALL-E irashobora gukora zeru-shoti ijwi rya clone uhereye hafi ya 3-isegonda ya sample ya disikuru itagaragara.
Ni ubuhe bwoko bwa codec ya VALL-E ikoresha kugirango ikore amajwi yayo?
VALL-E yubaka kuri __AIU_PROTECTED_13 __ 'EnCodec, ihanura ibimenyetso byayo bya acoustic kugirango bihuze imvugo.
Kuki VALL-E ikoresha byombi autoregressive hamwe na non-autoregressive stage?
Ikimenyetso cya Codec ni urwego runyuze kuri RVQ; VALL-E ihanura umugezi wa mbere wuzuye utabishaka kandi ibimenyetso bisigaye byerekana muburyo bwo gukora neza.
Hafi yamakuru angahe yo kuvuga VALL-E yatojwe, bigafasha gukomera kwa zeru?
VALL-E yatojwe amasaha agera ku 60.000 yo kuvuga, irenze kure imibare isanzwe ya TTS, yazamuye zeru-shusho.