Vokoder bu sukkandiko ci debit WaveGlow
WaveGlow vocodeur neuronal la bu bawoo ci NVIDIA, muy boole forme onde yu bawoo ci spectrogram mel ci benn yoon kese te du am benn régression.
Résumé
It matters because it delivers high-quality audio faster than real time using only a simple likelihood loss.
Plongeur bu xóot
WaveGlow, bi Prenger, Valle, ak Catanzaro génne ci NVIDIA ci 2018, dafa boole xalaat yu bawoo ci Glow ak WaveNet ngir defar vocoder bu gaaw te yomb tàggat. Vocoder GAN wuute na ak vocoder yi, ndax dafay normalise : dafay jàng benn kart buñ mëna soppi diggante distribution Gaussienne bu yomb ak forme onde audio, te dafa aju ci mel-spectrogram bi. Taggat yaram dafay yokk log-likelihood bi ci done yi, kon soxlawul benn diskriminatër bu wuute, amul benn auto-regression, te amul benn distillation bu am ñaari reso jàngalekat-ndongo bi njëkka am ci WaveNet parallèle. Ngir defar audio dangay jël misaalu bruit Gaussian ba noppi nga doxal reso buñ mëna soppi ci anam wu dellu ginaaw. WaveGlow dafay defar kàddu yu am kalite bu méngoo ak WaveNet ci noonu muy synthesize lu gëna gaaw ci waxtu dëgg ci GPU bu bees bi.
Gis-gis xarala
WaveGlow dafay jël jéego yuñ mëna soppi, bu nekk ci ñoom boole ci couche couplage affine ak benn convolution 1x1 buñ mëna soppi buñ leble ci Glow. Misali audio yi dañu leen di boole ci ay vecteur jaaraleko ci ab operaasioŋ squeeze suko defee ay couche couplage mëna leen soppi ci anam wu jaar yoon. Ndax jéego bu nekk mën nañu ko soppi, yoon wi jëm kanam dafay xayma limu mëna am ci tàggat yaram, yoon wi jëm ginaaw dafay méngale bruit bi ak audio bi ngir mëna am inference. Benn reso ak benn mébet log-likelihood bu baaxul mooy tax tàggat yaram gëna dëgër te yomb.
njeextalu pexe
Dugg ak yegg
Dafay gëna yombal jëfandikoo gi jaaraleko ci transkripsioŋ, nettali ak interfaasu baat.
Njëgg ak budget
Ekipu mejaa yi mën nañu yónnee audio bu leer ci anam wu gëna gaaw te seen xaalis gëna néew.
Gaawaay ak yaatuwaay
Sistem yiy jàkkarloo ak kiliyaan bi mën nañu def waxtaan ci anam wu gëna yaatu.
Ëlëgu Vocoder bu sukkandiko ci WaveGlow
WaveGlow dafa wane ni vocoder yu sell yi mën nañu xëcc kalite autoregressive, di indi jafe-jafe ci model audio yi ci topp. Yombaay bimu ñàkk benn yoon mingi wéy di xëcc nit ñi, waaye vocoders GAN yu melni HiFi-GAN leegi dañuy faral di am ndam ci dayo ak gaawaay. Bu ñu xoolee ci kanam, xalaat yu lalu ci debit ak debit-matching ñu ngi dellu ci TTS yu bees yi ci wetu diffusion, ak WaveGlow-style invertible designs ñu ngi wéy di joxe leeral ci gëstu ci njub, mëna saytu, ak defar forme vague bu baax.
Doxal ci àdduna dëgg
boole ak Tacotron 2 ci gasoduc TTS bu NVIDIA ngir defar kàddu yu am kalite studio
Synthese kàddu GPU bu gaaw ngir nettali, dubbing ak defar ëmbiit
Defar tàggat ak demo audio ci gëstu fu tàggat yaram bu amul benn ñàkk moo taamu
Baatu génnekaay bu mëna génne ci jamono dëgg ci sistem yiy weccoo xalaat yuy dawal ci kaw aparey NVIDIA
Risk yi ak balustrade yi
Jëfandikoo baat ci anam wu jaarul yoon ak niru ak nit dafay gëna yokk sudee nanguwul.
Jaar-jaar mën na wàññeeku ci aksan yi, dialect yi wala barab yu bari xumbaay.
Audio synthetik mën nañu ko jaawale ak wax ju dëggu sudee amul etiket bu leer.
Roadmap ngir samp gi
Wutal ndigal bu leer ngir jàpp baat bi, klone ko ak jëfandikoowaat ko.
Saytu kalite ci kàddukat yu bari ak anam yu bari ci ginaaw.
Mandargal kañ la nit wara xoolaat wala nangu ay génne.
Etiketu audio synthetik te nga denc dokimaa ci fimu bawoo ngir mëna lim.
Weyal di banneexu
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 WaveGlow Flow-Based 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
Gis bi ci topp
Voicebox Flow buy méngoo ak defar kàddu
Laaj yi ñuy faral di laaj
What is WaveGlow Flow-Based Vocoder?
WaveGlow vocodeur neuronal la bu bawoo ci NVIDIA, muy boole forme onde yu bawoo ci spectrogram mel ci benn yoon kese te du am benn régression. Dafa am solo ndax dafay joxe audio bu baax te gëna gaaw ci waxtu dëgg, jëfandikoo lu yomb lu mëna ñàkk.
WaveGlow dafa boole xalaati architecture ci ban xeetu architecture?
WaveGlow dafay boole xeetu debit buñ mëna soppi bu Glow ak jëmmal modelu audio bu WaveNet.
Ban xeetu model mooy WaveGlow?
WaveGlow ab debit buy normalisasioŋ la buy jàng xayma buñu mëna soppi diggante bruit Gaussian ak audio.
naka la ko WaveGlow di defaree forme vague ci waxtu inférence bi?
Ndax reso bi mën nañu ko soppi, dangay génne bruit Gaussian ba noppi nga yóbbu debit bi ci ginaaw, lépp di aju ci mel-spectrogram bi.
Ban mébet la WaveGlow di gëna mëna jëfandikoo ci diiru tàggat yaram?
Bi muy daw, WaveGlow dafay yokk limu mëna am ci log bi, te soxlawul benn tànneef wala distilaasioŋ.
Yan ñaari mbir ñoo bokk ci jéego bu nekk ci WaveGlow?
Jéego bu nekk dafay boole ab couplage affine ak ab convolution 1x1 buñ mëna soppi buñu jëlee ci Glow.