Isbarbardhigga WaveGAN Vocoder
Parallel WaveGAN waa cod-bixiye neural ah oo degdeg ah oo u beddela mel-spectrogram qaab mowjad maqal ah oo cayriin ah iyadoo la adeegsanayo GAN yar, oo soo saaraysa dhammaan muunadaha hal mar.
Dulmar
It matters because it gives near-real-time, high-quality speech with a compact model.
quusid qoto dheer
Codkadeeyuhu waa heerka ugu dambeeya ee dhuumaha TTS: waxay u beddeshaa khariidad muuqaal dhawaaqeed ah (badanaa mel-spectrogram) oo u beddela hirarka dhawaaqa dhabta ah ee aad maqashid. Isbarbardhigga WaveGAN, oo ay soo jeediyeen Yamamoto, Song, iyo Kim sanadka 2019, ayaa tan ku sameeya koronto-dhaliye qaab WaveNet-ka aan autoregressive-ka ahayn oo loo tababaray sidii shabakad iska soo horjeeda. Halkii la saadaalin lahaa hal muunad maqal ah sida WaveNet-kii asalka ahaa, waxay soo saartaa qaab-dhismeedka hirarka oo dhan si barbar socda, taasoo ka dhigaysa mid aad u degdeg badan. Cuntodeeda muhiimka ah waxay ku daraysaa khasaare iska soo horjeeda oo leh xalal badan oo waqti gaaban ah oo luminaya Fourier transform (STFT), markaa moodeelku wuxuu u dhigmaa calaamadda dhabta ah ee waqtiyo badan iyo miisaan joogto ah. Natiijadu waa koronto-dhaliye yar (qiyaastii 1.4 milyan oo cabbir) kaas oo ku shaqeeya marar badan ka dhaqso badan waqtiga dhabta ah ee GPU-ga.
Aragtida Farsamada
Koronto-dhaliye waa shabakad fidsan-convolution oo ku xidhan mel-spectrogram iyo galinta qaylada, qaylada samaynta iyo sifooyin toos ah muunadaha. Tababarka wadajirka ah wuxuu yareeyaa lumitaanka STFT ee xallinta badan, oo lagu xisaabiyo isbarbardhigga cabbirrada cabbirka ee dhowr cabbir FFT ah iyo dhererka dhererka, iyo khasaaraha iska soo horjeeda ee ka imanaya takoore xukumaya xaqiiqada. Erayga STFT waxa uu dejiyaa oo uu dedejiyaa tababbarka iska hor imaadka, isaga oo qabsanaya tafaasiil fiican iyo qaab muuqaal ballaadhan oo aan la kala saarin.
Saamaynta Istiraatijiyadeed
Helitaanka iyo gaarsiinta
Waxay wanaajisaa marin u helida iyada oo loo marayo qoraal-qorid, sheeko, iyo is-dhexgalyo cod.
Qiimaha iyo miisaaniyada
Kooxaha warbaahintu waxay ku soo rari karaan codka sifaysan si degdeg ah iyagoo wata miisaaniyado yaryar.
Xawaaraha iyo miisaanka
Nidaamyada u jeedda macmiisha waxay ka baaraandegi karaan isdhexgalka hadalka si weyn.
Mustaqbalka Isbarbardhiga WaveGAN Vocoder
Isbarbardhigga WaveGAN waxay gacan ka geysatey sameynta cod-bixiyayaasha GAN inay yihiin kuwa wax ku ool ah, iyo luminteeda xallinta badan ee STFT hadda waxay u muuqataa dhammaan kuwa guulaysta sida HiFi-GAN iyo habab badan oo qulqulka. Mawduucani waxa uu tilmaamayaa cod-bixiyeyaal yar-yar, cod-bixiyeyaal-hooseeya oo loogu talagalay kaaliyeyaasha qalabka, aalada maqalka, iyo beddelka codka nool, iyo cod-bixiyayaasha caalamiga ah ee guud ahaan ku hadla kuwa aan la arkin. Filo is dhexgal adag oo leh dhamaadka-ilaa-dhamaadka TTS iyo hawlgelinta hufan ee mobilada iyo chips-ka ku xidhan.
Dhaqangelinta Adduunka-dhabta ah
Soo saarida hadalka-waqtiga-dhabta ah ee kaaliyayaasha codka mobaylka halka daahida iyo cabbirka qaabka ay muhiim tahay
U adeegida koronto-dhaliyaha waveform-ka oo ay weheliyaan moodooyinka dhawaaqa sida Tacotron 2 ama FastSpeech
Qalab-ku-hadal-ku-hadal qalabyada gelitaanka ee aan isku hallayn karin daruuraha
Hababka beddelka codka ee dib u soo saara spectrogram-yada loo beddelay maqal-dhawaaq dabiici ah
Khatarta & Dariiqyada Ilaalada
Si xun u isticmaalka codka iyo khataraha is-yeelyeelku way kordhaan marka oggolaanshaha la waayo.
Saxnimadu waxay hoos ugu dhici kartaa lahjadaha, lahjadaha, ama jawiga buuqa badan.
Maqalka synthetic waxaa lagu khaldi karaa hadal dhab ah iyada oo aan si cad loo calaamadin.
Qorshe Hawleedka Dhaqangelinta
Hel ogolaansho cad oo ku saabsan qabashada codka, xidhitaanka, iyo dib u isticmaalka
Tijaabi tayada ku hadasha kala duwan iyo xaaladaha asalka.
Qeex marka bani'aadamku ay tahay inuu dib u eego ama oggolaado wax soo saarka.
Ku calaamadee codka synthetic oo xafid diiwaannada la-xisaabtanka.
Sii wad Sahaminta
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 Parallel WaveGAN Vocoder 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
Hagaha xiga
Codka-dhaliyaha MelGAN
Su'aalaha soo noqnoqda
What is Parallel WaveGAN Vocoder?
Parallel WaveGAN waa cod-bixiye neural ah oo degdeg ah oo u beddela mel-spectrogram qaab mowjad maqal ah oo cayriin ah iyadoo la adeegsanayo GAN yar, oo soo saaraysa dhammaan muunadaha hal mar. Waa muhiim sababtoo ah waxay ku siinaysaa wakhtiga-dhabta ah, hadal tayo sare leh oo leh qaab is haysta.
Waa maxay shaqada cod-bixiyaha sida Parallel WaveGAN?
Codkadeeyuhu waa marxaladda ugu dambeysa ee TTS ee ka soo saarta mowjadda dhawaaqa dhabta ah ee sifooyinka dhawaaqa sida mel-spectrogram.
Sidee buu barbar socda WaveGAN u soo saaraa muunado maqal ah marka la barbar dhigo WaveNet asalka ah?
Isku midka ah WaveGAN waa mid aan autoregressive ahayn, oo soo saaraya qaab mawjada oo dhan hal mar halkii muunad laga qaadan lahaa, taasoo ka dhigaysa mid aad u dhakhso badan.
Khasaare kee baa udub dhexaad u ah isbarbardhigga WaveGAN marka laga reebo khasaaraha iska soo horjeeda?
Waxay isku daraysaa khasaare iska soo horjeeda iyo lumin xalal badan oo STFT ah kaas oo isbarbar dhigaya cabbirka cabbirka cabbirrada kala duwan.
Waa maxay qaab-dhismeedka uu isticmaalo koronto dhaliyaha WaveGAN?
Koronto-dhaliye waa shabakad u eeg WaveNet oo laga dhisay is-faham-la'aan, oo ku xidhan mel-spectrogram iyo buuq.
Waa maxay sababta isbarbardhigga WaveGAN u soo jiidanayso keenista?
Iyada oo ku dhawaad 1.4 milyan oo xuduudo ah ay yar tahay oo ay marar badan ka dhaqso badan tahay waqtiga dhabta ah, kuna habboon isticmaalka aaladda.