I-UnivNet Multi-Resolution Vocoder
I-UnivNet iyivokhoda ye-GAN ehlulelayo ekhiqize umsindo kusetshenziswa ama-spectrogram amaningi enziwe ngekhompyutha ngezinqumo ezihlukene ze-STFT, elola imininingwane yefrikhwensi ephezulu.
Uhlolojikelele
It aims to be a universal vocoder that generalizes well to unseen speakers and recording conditions.
I-Deep Dive
I-UnivNet, ihlongozwe nguJang et al. ngo-2021, ibhekana nobuthakathaka obujwayelekile kumavokhoda e-GAN: amafrikhwensi aphezulu ahlanganisiwe noma agcwele i-artifact. Izimo zayo zokuphehla ugesi ku-full-band mel-spectrograms futhi isebenzisa i-location-variable convolutions (LVC), lapho ama-convolution kernels ebikezelwa khona ngokuhamba esuka kuzici zokufaka ukuze isihlungi sivumelane nokuqukethwe kwendawo. Umbono wesihloko i-multi-resolution spectrogram discriminator (MRSD): esikhundleni sokwahlulela kuphela i-waveform eluhlaza, i-UnivNet ihlanganisa ama-STFT amaningana ngamawindi ahlukene nosayizi be-hop futhi isebenzisa ababandlululi kulawo magnitude e-spectrogram. Lokhu kuphusha ijeneretha ukuthi ithole kokubili imininingwane emihle ye-spectral kanye nesakhiwo sesikhashana esibanzi esifanele. Iqeqeshelwe izikhulumi eziningi, i-UnivNet ikhiqiza inkulumo yemvelo yamazwi engakaze iwabone ngesikhathi sokuqeqeshwa, izuza ilebula yayo yomhlaba wonke.
I-Technical Insight
I-convolution ye-UnivNet eguquguqukayo yendawo ikhiqiza izisindo zayo ze-kernel ngokuguquguqukayo kusukela kuzici ze-conditioning mel ngenethiwekhi encane ye-kernel-predictor, ngakho isikhathi ngasinye isinyathelo sisebenzisa ngempumelelo isihlungi esivumelana nezimo kune-kernel eyabiwe engaguquki. Kuhlanganiswe ne-multi-resolution spectrogram discriminator, ehlanganisa ukuhwebelana kwezikhathi ezimbalwa ngasikhathi sinye, lokhu kuqondise ngokuqondile ibhendi yamafrikhwensi aphezulu lapho amavokhoda alula we-GAN evamise ukufiphala noma ukuduma.
I-Strategic Impact
Finyelela futhi ufinyelele
Ithuthukisa ukufinyeleleka ngokuloba, ukulandisa, nezixhumi ezibonakalayo zezwi.
Izindleko kanye nesabelomali
Amaqembu emidiya angathumela umsindo opholishiwe ngokushesha ngamabhajethi amancane.
Isivinini nesikali
Amasistimu abhekene nekhasimende angacubungula ukusebenzelana okukhulunyiwe ngesilinganiso esikhulu.
Ikusasa le-UnivNet Multi-Resolution Vocoder
Ukubandlululwa kwe-spectrogram ye-UnivNet enezixazululo eziningi sekuyisithako esijwayelekile kuzitaki zesimanje ze-TTS namasistimu athonyekile afana ne-BigVGAN namakhodekhi omsindo we-neural. Lindela uhlaka lwendawo yonke, lwesipikha lwe-agnostic ukuze luqhubeke lunwebeka luye ezwini eliculayo, ukuhlanganiswa kwezilimi eziningi, nomsindo womkhawulokudonsa ogcwele ongu-48 kHz, kuyilapho umbono we-adaptive-kernel wazisa amamodeli akudivayisi asebenza kahle okumele aphathe amazwi ahlukahlukene ngaphandle kokucushwa kahle kwesikhulumi ngasinye.
Ukuqaliswa Komhlaba Wangempela
Amasevisi we-TTS wezipikha eziningi okufanele azwakale engokwemvelo emazwini angekho kudatha yokuqeqeshwa
Amapayipi wokuhlanganisa izwi lapho ivowuda eyodwa yendawo yonke inikezela ngezipikha eziqondiwe
I-high-fidelity audiobook nokulandisa kwe-podcast okudinga ukufana okupholile kanye namafrikhwensi aphezulu
I-backend vocoder yezinhlelo ze-TTS eziya ekupheleni ezibhanqa i-spectrogram predictor ne-robust waveform generator
Izingozi & Guardrails
Ukusetshenziswa kabi kwezwi kanye nezingozi zokuzenza ongeyena ziyanda uma imvume ingekho.
Ukunemba kungase kwehle kuzo zonke izinhlobo zokuphimisela, izilimi zesigodi, noma izindawo ezinomsindo.
Umsindo wokwenziwa ungenziwa iphutha njengenkulumo eyiqiniso ngaphandle kokulebula okucacile.
Ukuqalisa Umhlahlandlela
Thola imvume esobala yokuthwebula izwi, ukuhlanganisa, nokusebenzisa kabusha.
Ikhwalithi yokuhlola kuzo zonke izipikha nezimo zangemuva.
Chaza ukuthi kunini lapho umuntu kufanele abuyekeze noma agunyaze okuphumayo.
Lebula umsindo wokwenziwa futhi ugcine amarekhodi atholakalayo ukuze aziphendulele.
Qhubeka Uhlole
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 UnivNet Multi-Resolution 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
Umhlahlandlela olandelayo
Ama-Neural Vocoders
Imibuzo evame ukubuzwa
What is UnivNet Multi-Resolution Vocoder?
I-UnivNet iyivokhoda ye-GAN ehlulelayo ekhiqize umsindo kusetshenziswa ama-spectrogram amaningi enziwe ngekhompyutha ngezinqumo ezihlukene ze-STFT, elola imininingwane yefrikhwensi ephezulu. Ihlose ukuba i-vokhoda yendawo yonke ehlanganisa kahle izikhulumi ezingabonwa nezimo zokuqopha.
Siyini isici sokusayina sobandlululo lwe-UnivNet?
I-UnivNet's multi-resolution spectrogram discriminator ihlaziya ama-STFT amaningana ngamawindi ahlukene nosayizi be-hop ukuze ithwebule ukuhweba okuhlukile kwemvamisa yesikhathi.
Iyiphi inkinga kumavokhoda angaphambili we-GAN i-UnivNet eqondise ngqo kuwo?
Ngokubandlulula kuzo zonke izinqumo ze-spectrogram eziningi, i-UnivNet ithuthukisa imininingwane yamafrikhwensi aphezulu amavokhoda alula we-GAN avame ukuwufiphalisa.
Yiziphi i-convolutions yendawo eguquguqukayo (LVC) ku-UnivNet?
I-LVC isebenzisa inethiwekhi ye-kernel-predictor ngakho indawo ngayinye isebenzisa izihlungi eziguquguqukayo zokuqukethwe ezithathwe ku-conditioning mel-spectrogram.
Kungani i-UnivNet ichazwa njengevokhoda yendawo yonke?
Iqeqeshelwe izipikha eziningi, i-UnivNet ihlanganisa inkulumo yemvelo ngisho namazwi engakaze ihlangane nawo ekuqeqeshweni, yingakho itholakala emhlabeni wonke.
Ingabe i-multi-resolution spectrogram discriminator isiza kanjani ijeneretha?
Izinqumo ezihlukile ze-STFT zigcizelela ukuhwebelana okuhlukile kwemvamisa yesikhathi, ngakho ukufaniswa kwakho konke kuphushela ijeneretha emininingwaneni nesakhiwo esinembile.