Ụdị Asụsụ VALL-E na Codec
VALL-E megharịrị ederede-gaa-okwu dị ka nsogbu nhazi asụsụ n'elu akara codec ọdịyo, na-eme ka mmechi olu site na naanị sekọnd atọ nke nlele.
Nchịkọta
It showed that the same next-token prediction powering text LLMs can generate remarkably natural, expressive speech.
Ime miri emi
Nke Microsoft mara ọkwa na mbido 2023, VALL-E na-ewere nhazi okwu dịka nhazi asụsụ. Kama ibu amụma spectrogram, ọ na-ebu amụma akara ụda olu pụrụ iche nke codec neural (EnCodec), yabụ ọgbọ na-aghọ amụma na-esote akara maka okwu ọdịyo. Nyere ndekọ nke abụọ nke abụọ nke ọkà okwu a na-adịghị ahụ anya gbakwunyere ederede ebumnuche, VALL-E na-aga n'ihu n'olu ọkà okwu ahụ, na-echekwa timbre na ọbụna gburugburu ụda olu. A zụrụ ya n'ihe dị ka awa 60,000 nke ikwu okwu, nke ukwuu karịa ụdị datasets TTS, nke nyere ya cloning na-enweghị atụ. N'ihi codec tokens na-layeded (site na RVQ), VALL-E na-eji abụọ nkebi: autoregressive nlereanya na-ebu amụma nke mbụ, keadighi rarii token iyi oyi na ozugbo, na-abụghị autoregressive nlereanya na-ejuputa na fọdụrụ nkọwa token. Ntụziaka codec-LM a kpaliri ndị ga-anọchi anya VALL-E 2 na ọtụtụ ụdị ntọala okwu.
Nghọta nka nka
Aghụghọ a bụ ngbanwe ngwakọ n'elu akara codecs hierarchical. The autoregressive ogbo na-ebu amụma ndị kasị mkpa kasị mkpa-codebook token otu n'otu oge, na-ewere prosody na ọdịnaya. A na-ebu amụma akwụkwọ koodu ndị fọdụrụnụ, bụ́ ndị na-agbakwụnye nkọwapụta ụda dị mma, n'otu aka ahụ site na ihe nlere na-abụghị nke autoregressive kwadoro na iyi nke mbụ yana ngwa ngwa ọkà okwu. Nkewa nke a na-eme ka ịdịmma ya dị elu ma na-ezere ọnụ ahịa nke imepụta akara ngosi ọ bụla n'usoro, yana iji codec pụtara enwere ike iji otu igwe ihe ngbanwe were were ụdị okwu na ederede.
Mmetụta atụmatụ
Nweta na iru
Ọ na-eme ka nnweta ya dịkwuo mma site na ndegharị, ịkọ akụkọ, na ntụgharị olu.
Ọnụ ego na mmefu ego
Ndị otu mgbasa ozi nwere ike ibubata ọdịyo a na-egbu maramara ngwa ngwa site na iji obere mmefu ego.
Ọsọ na ọnụ ọgụgụ
Sistemụ na-eche ihu ndị ahịa nwere ike hazie mkparịta ụka n'ọtụtụ buru ibu.
Ọdịnihu nke VALL-E na Ụdị Asụsụ Codec
Ụdị asụsụ Codec na-ejikọta okwu na ụdị asụsụ buru ibu, na-atụ aka na sistemu ejikọtara ọnụ na-ege ntị, na-atụgharị uche na na-ekwu okwu n'otu ụdị. Na-atụ anya nkwụsi ike ka mma na obere arịa dị ole na ole, ọgbọ na-enuba n'ezie, yana njikwa ike karịa mmetụta na ụdị. Otu cloning siri ike nke na-eme VALL-E bara uru maka ịnweta na dubbing na-ebulikwa nchegbu miri emi na nkwenye, ya mere akara mmiri, nchekwa nkwenye olu, na nchekwa amụma na-aghọ akụkụ bụ isi nke otu esi ebufe usoro ndị a.
Mmejuputa n'ezie n'ụwa
Imechi olu site na sekọnd ole na ole nke ọdịyo maka ndị enyemaka ahaziri ahazi ma ọ bụ ngwaọrụ nnweta na-eweghachi olu furu efu
Ịkpọsa na ịtụgharị vidiyo n'asụsụ ndị ọzọ ka ị na-edobe timbre nke ọkà okwu mbụ
Na-ewepụta akụkọ nkwuwapụta, dabara na okirikiri nke na-echekwa gburugburu ụda egwu ndekọ
Na-eje ozi dị ka ọkpụkpụ azụ okwu na ndị enyemaka multimodal na-aghọta ma na-emepụta ọdịyo a na-ekwu
Ihe ize ndụ & okporo ụzọ nche
Iji olu eme ihe na ihe egwu mpụta ga-abawanye mgbe nkwenye na-efu.
Izi ezi nwere ike ịdaba n'ofe ụda olu, olumba ma ọ bụ gburugburu mkpọtụ.
Enwere ike imehie ọdịyo sịntetik dị ka ezigbo okwu na-enweghị akara doro anya.
Map mmejuputa
Nweta nkwenye doro anya maka ijide olu, imechi, na ijigharị.
Nwale ogo n'ofe ndị na-ekwu okwu dị iche iche yana ọnọdụ ndabere.
Kọwaa mgbe mmadụ ga-enyocha ma ọ bụ kwado nsonye.
Deba aha ọdịyo sịntetik ma debe ndekọ ihe ndekọ maka ịza ajụjụ.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Ikike nke Ụdị Asụsụ buru ibu
Ajụjụ a na-ajụkarị
What is VALL-E and Codec Language Models?
VALL-E megharịrị ederede-gaa-okwu dị ka nsogbu nhazi asụsụ n'elu akara codec ọdịyo, na-eme ka mmechi olu site na naanị sekọnd atọ nke nlele. O gosiri na otu amụma amụma na-esote na-enye ike ederede LLM nwere ike iwepụta okwu ebumpụta ụwa n'ụzọ pụtara ìhè.
Kedu isi echiche na-ekewa VALL-E na sistemu ederede gaa na okwu mbụ?
VALL-E na-emegharị TTS dị ka amụma akara na-esote maka akara codec ọdịyo dị iche, na-ewere ọgbọ okwu dị ka ụdị asụsụ.
Ego ole nrụtụ aka ka VALL-E chọrọ iji mechie olu ọkà okwu ọhụrụ?
VALL-E nwere ike ịme mmechi olu efu site na ihe dị ka ihe dị ka sekọnd atọ nke ọkà okwu a na-adịghị ahụ anya.
Kedu codec neural ka VALL-E na-eji wepụta akara ọdịyo ya?
VALL-E na-ewuli na Meta's EnCodec, na-ebu amụma akara ụda olu ya pụrụ iche iji mepụta okwu.
Kedu ihe kpatara VALL-E ji eji ma autoregressive na nke na-abụghị autoregressive ogbo?
Codec token bụ usoro nhazi site na RVQ; VALL-E na-ebu amụma iyi mmiri nke mbụ na-agba ọsọ na-akpaghị aka yana nkọwa nkọwa ndị fọdụrụ n'otu n'otu maka ịrụ ọrụ.
Ihe dị ka ole data okwu ka a zụrụ VALL-E na, na-eme ka cloning na-agba ọsọ siri ike?
A zụrụ VALL-E n'ihe dị ka awa 60,000 nke ikwu okwu, karịa ihe ndekọ data TTS ahụkarị, bụ nke kwalitere nchikota efu ya.