Ụdị mgbasa ozi maka ọdịyo
Ụdị mgbasa ozi na-emepụta ọdịyo site n'ịmụ ịtụgharị usoro mkpọtụ nzọụkwụ site na nzọụkwụ, na-atụgharị mkpọtụ enweghị usoro n'ime okwu ọnụ, egwu, ma ọ bụ mmetụta ụda.
Nchịkọta
They power many of today's most realistic text-to-audio and music-generation systems.
Ime miri emi
Ụdị mgbasa ozi maka ọdịyo na-agbaziri otu echiche bụ isi nke gbanwere ọgbọ onyonyo. N'oge ọzụzụ, a na-eji nwayọọ nwayọọ mebie ọdịyo dị ọcha site n'ịgbakwunye ụda Gaussian n'ọtụtụ usoro ruo mgbe ọ ga-adị ọcha. Netwọk akwara na-amụta ịkọ ma wepụ mkpọtụ ahụ na nzọụkwụ ọ bụla. N'oge ọgbọ, ihe nlereanya ahụ na-amalite site na mkpọtụ na-enweghị ihe ọ bụla ma na-ekwu okwu ugboro ugboro, site na ngwa ngwa ederede na-eduzi ya, iji mepụta mgbama dị ọcha. Ọtụtụ sistemu na-arụ ọrụ ọ bụghị n'ụdị ebili mmiri dị ndụ kama na nhụsianya latent abịakọrọ ma ọ bụ spectrogram, nke na-eme ka ọgbọ na-adị ngwa ngwa ma nwee ike ịnweta ya. Ihe atụ ama ama gụnyere AudioLDM, Stable Audio, na Riffusion. Nsonaazụ bụ nkwesị ntụkwasị obi dị elu, njikọ ọdịyo a na-achịkwa n'ofe okwu, egwu, na ụda gburugburu ebe obibi.
Nghọta nka nka
Kama iwepụta ụdị ebili mmiri dị ogologo ozugbo, ọtụtụ ụdị mgbasa ozi ọdịyo na-arụ ọrụ n'ime oghere mmụta mmụta nke ihe ngbanwe autoencoder mepụtara, ma ọ bụ na mel-spectrograms mechara gbanwee ka ọ bụrụ ụda site na ụda olu dị ka HiFi-GAN. A na-agbanye ntụnye ederede site na nlebara anya, na-ejikarị ntinye CLAP na-ahazi ọdịyo na asụsụ. A na-eji usoro dị ka DDIM na distillation kwalite ọsọ nlere anya, na-egbutu narị otu narị nzọụkwụ na-emebi emebi ruo naanị njuaka.
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 ụdị mgbasa ozi maka ọdịyo
Na-atụ anya nleba anya ngwa ngwa site na ụdị na-agbanwe agbanwe na distillation, na-aga n'ihu n'ọgbọ dị adị na mgbasa ozi. Ogologo oge, ihe egwu egwu ahaziri ahazi nwere ịdị n'otu n'amaokwu-ịkwe na-apụta, n'akụkụ njikwa kacha mma site na ịse, ị ga na ọdịyo ntụaka. Usoro multimodal nke jikọtara ọnụ na-emepụta vidiyo na ụda ụda mekọrịtara ọnụ na-aga n'ihu ngwa ngwa. Ka ịdịmma na-arị elu, akara mmiri na ngwaọrụ ihe eji eme ihe ga-adị mkpa iji leba anya na adịgboroja miri emi, ịkpọchi olu, na nchegbu ikike nwebisiinka egwu.
Mmejuputa n'ezie n'ụwa
Stable Audio na-emepụta egwu ndabere na-enweghị eze na mmetụta ụda sitere na ngwa ngwa ederede maka ndị okike vidiyo
AudioLDM na-emepụta ụda gburugburu ebe obibi dị ka mmiri ozuzo, nzọụkwụ ụkwụ, ma ọ bụ nkịta na-agbọ maka egwuregwu na ihe nkiri foley
Riffusion na-emepụta obere egwu egwu site na ịkatọ onyonyo spectrogram nwere ọnọdụ na ụdị na mkpali ngwa.
Sistemu ederede gaa na-okwu dabere na mgbasa ozi na-emepụta akụkọ sitere n'okike, na-ekwupụta maka akwụkwọ ọdịyo na ndị enyemaka olu.
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 Diffusion Models for Audio 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
Bark Generative Audio Model
Ajụjụ a na-ajụkarị
What is Diffusion Models for Audio?
Ụdị mgbasa ozi na-emepụta ọdịyo site n'ịmụ ịtụgharị usoro mkpọtụ nzọụkwụ site na nzọụkwụ, na-atụgharị mkpọtụ enweghị usoro n'ime okwu ọnụ, egwu, ma ọ bụ mmetụta ụda. Ha na-eme ka ọtụtụ n'ime usoro ederede gaa na ọdịyo na usoro ọgbọ egwu kachasị taa.
Kedu usoro bụ isi ihe nlereanya mgbasa ozi na-amụta n'oge ọzụzụ?
A zụrụ ụdị mgbasa ozi ịkọ amụma na iwepu ụda Gaussian agbakwunyere na nzọụkwụ ọ bụla, ka ha wee mechaa gbanwee mkpọtụ dị ọcha ka ọ bụrụ ọdịyo dị ọcha.
Kedu ihe kpatara na ọtụtụ ụdị mgbasa ozi ọdịyo na-arụ ọrụ na latent ma ọ bụ spectrogram kama ịhịa aka n'ahụ ebili mmiri?
Ịrụ ọrụ n'ime oghere a gbakọrọ agbakọ ma ọ bụ na spectrogram na-ebelata oke akụkụ, na-eme ọzụzụ na nlele anya nke ọma karịa ịmegharị ogologo ụdị ebili mmiri ozugbo.
Kedu otu esi eji ngwa ngwa ederede eduzi ọgbọ ọdịyo n'ụdị ndị a?
A na-enyekarị nkwado ederede n'ime netwọk ndị na-eme mkpesa site na nlebara anya, na-ejikarị ihe ntinye CLAP na-ahazi asụsụ na ọdịyo.
Mgbe a na-emepụta spectrogram, kedu ka esi atụgharịghachi ya azụ ka ọ bụrụ ụda?
Vocoder dị ka HiFi-GAN na-atụgharị mel-spectrogram emepụtara ka ọ bụrụ ụdị ụda a na-anụ.
Kedu usoro na-enyere aka mee ka ọgbọ dị ngwa site n'ibelata ọnụ ọgụgụ nke usoro nkwụsị?
Ụdị nhụsianya na nkwụsi ike na-akpakọba ọtụtụ narị usoro nkwubi okwu n'ime ole na ole, na-eme ka ọ dị ngwa ngwa, nke nwere ike ịlele oge.