Ntuziaka Audio AI

HiFi-GAN na GAN Vocoders

HiFi-GAN bụ ụda olu na-eme ka onye iro na-atụgharị mel-spectrogram ka ọ bụrụ usoro ebili mmiri na-adịghị ahụkebe ihe fọrọ nke nta ka ọ bụrụ ozugbo, na-emepụta okwu dị mma nke studio ngwa ngwa karịa ozugbo.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

Ọ ghọrọ ọkwa ikpeazụ nke ederede nke oge a n'ihi na ọ dị ngwa ngwa, dị fechaa, ma sie ike ịmata ọdịiche dị na ezigbo ndekọ.

Ime miri emi

Vocoder bụ nzọụkwụ ikpeazụ n'ime ọtụtụ pipeline TTS: ihe nlereanya dị ka Tacotron ma ọ bụ FastSpeech na-ebu amụma mel-spectrogram (eserese foto nke ugboro ugboro n'oge), na ụda olu na-ejupụta n'ezie ụdị ebili mmiri. Vocoders mbụ akwara dị ka WaveNet dara ụda nke ukwuu mana ewepụtara ihe onyonyo site na nlele, na-eme ka ha dị nwayọ nwayọ. HiFi-GAN, nke Kong, Kim na Bae weputara na 2020, jiri otu onye na-ebupụta ihe na-ebuga azụ azụ azụ n'agha dochie akaghị aka ahụ. Ihe aghụghọ ya bụ iji ọtụtụ ịkpa ókè na-ekpe ikpe ọdịyo n'ụdị dị iche iche na usoro oge dị iche iche, na-amanye onye na-emepụta ọkụ ka ọ nweta ma ọdịdị dị mma na oge pitch ziri ezi. Nsonaazụ bụ okwu 22 kHz chịkọtara ọtụtụ narị ugboro ngwa ngwa karịa ezigbo oge na GPU, yana ọdịyo na-ama nke ọma n'ala-eziokwu.

Nghọta nka nka

Onye na-emepụta ihe nke HiFi-GAN na-ebuli mel-spectrogram ahụ site na ntụgharị ntụgharị, yana ihe mgbochi Multi-Receptive Field nke na-agwakọta nha kernel dị iche iche na dilation iji weghara usoro ebili mmiri dị iche iche. Ezinụlọ abụọ na-akpa oke na-eme ndị uwe ojii: Onye ịkpa oke oge dị iche iche na-emegharị akara ngosi 1D n'ime grids 2D na primes dị ka 2, 3, 5, 7, 11 iji jide oge oge, na onye na-akpa ókè ọtụtụ na-enyocha ebili mmiri n'ọtụtụ mkpebi ala. Mel-spectrogram na mfu dakọtara na-eme ka ọzụzụ ahụ kwụsie ike.

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 HiFi-GAN na GAN Vocoders

Ndị vocoders GAN na-adịwanye nta na ngwa ngwa: ụmụ dị ka BigVGAN na-agbakwunye mmemme mgbochi iji gbasaa n'ofe ndị na-abụ abụ, ngwa na asụsụ a na-adịghị ahụ anya, ebe UnivNet na Vocos na-aga n'ihu n'ihu ọha, njikọ njikọ niile. Ụdị gụgharia na n'ime ngwaọrụ ugbu a na-agba ọsọ vocoding n'ime ekwentị yana ntị ntị maka ndị enyemaka obere oge. Na-abawanye ụba, mgbasa ozi na ụdị ọdịyo dabara adaba na-agbasa n'ime ụdị GAN na-emepụta otu ngafe, na-agwakọta ikwesị ntụkwasị obi mgbasa na ọsọ GAN. Na-atụ anya ka ndị vocoders ga-adaba n'ime codecs ọdịyo akwara na-eme ka okwu ma egwu dị.

Mmejuputa n'ezie n'ụwa

Na-emepụta mmepụta okwu nke ndị enyemaka mebere na ngwa nsoroụzọ chọrọ nzaghachi na-enweghị igbu oge.

Na-eme ka mmechi olu na-eme ihe n'ezie na ntinye akwụkwọ ebe a na-atụgharị mel-spectrogram cloned ka ọ bụrụ ọdịyo na-ada ụda.

Akwụkwọ ọdịyo na nyiwe ịkọ akụkọ pọdkastị na-emepụta awa okwu ọsọ ọsọ na ọnụ ala.

Na-eje ozi dị ka ogbo a na-efegharị efegharị n'ime ndị na-emepụta ụda olu na ngosi egwu site na ụdị ụda olu ụwa niile nke BigVGAN.

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

1

Nweta nkwenye doro anya maka ijide olu, imechi, na ijigharị.

2

Nwale ogo n'ofe ndị na-ekwu okwu dị iche iche yana ọnọdụ ndabere.

3

Kọwaa mgbe mmadụ ga-enyocha ma ọ bụ kwado nsonye.

4

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 HiFi-GAN and GAN Vocoders quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Malite ajụjụ

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ntuziaka na-esote

Mgbakwunye WaveGAN Vocoder

Ajụjụ a na-ajụkarị

Kedu ihe bụ HiFi-GAN na GAN Vocoders?

HiFi-GAN bụ ụda olu na-eme ka onye iro na-atụgharị mel-spectrogram ka ọ bụrụ usoro ebili mmiri na-adịghị ahụkebe ihe fọrọ nke nta ka ọ bụrụ ozugbo, na-emepụta okwu dị mma nke studio ngwa ngwa karịa ozugbo. Ọ ghọrọ ọkwa ikpeazụ n'ọkwa ikpeazụ nke ederede gaa n'okwu n'ihi na ọ na-adị ngwa ngwa, dị fechaa, na nke siri ike ịmata ọdịiche dị na ihe ndekọ dị adị.

Kedu ọrụ bụ isi nke ụda olu dị ka HiFi-GAN na ọkpọkọ ederede gaa na-okwu?

Vocoder bụ ogbo ikpeazụ nke na-emepụta ihe nlere anya na-efegharị efegharị n'ezie sitere na mel-spectrogram nke ụdị ụda olu mepụtara.

Kedu ihe kpatara HiFi-GAN ji dị ngwa karịa ụda olu WaveNet mbụ?

WaveNet mepụtara sample audio-site-sample (autoregressively), ebe onye na-ebubata nri HiFi-GAN na-ewepụta usoro ifegharị ahụ n'otu oge, na-enweta ọsọ ọsọ ọsọ ọsọ karịa oge.

Kedu ihe ịkpa oke oge ọtụtụ oge HiFi-GAN na-enyere aka ijide?

Onye ịkpa oke oge dị iche iche na-edogharị ụdị ebili mmiri 1D ka ọ bụrụ 2D site na iji oge ọnụọgụ dị mkpa iji chọpụta nhazi oge ejikọtara na pitch.

A zụrụ vocoders GAN 'na-emegide.' Kedu ihe nke ahụ pụtara ebe a?

Na ntọlite ​​​​GAN, onye na-emepụta ọkụ na-amụta imepụta ebili mmiri nke dabara adaba iji ghọgbuo netwọkụ ịkpa oke a zụrụ azụ ikwu eziokwu site na ọdịyo sịntetik.

Kedu ụda olu mechara gbasaa HiFi-GAN ka ọ chịkọta nke ọma ka ụda olu, abụ na ngwa anaghị ahụ anya?

BigVGAN na-ebuli HiFi-GAN ma na-agbakwụnye mmegide oge ọ bụla na-emegide ya, na-emeziwanye mkpokọta na ọdịyo na-enweghị nkesa dị ka abụ na ngwa.