HiFi-GAN ak GAN
HiFi-GAN vocoder buy génne ay noon la, muy soppi mel-spectrogram ci forme onde audio bu ñor ci saasi, di defar kàddu yu am kalite studio bu gëna gaaw ci jamono dëgg.
Résumé
It became the standard final stage of modern text-to-speech because it is fast, lightweight, and hard to distinguish from real recordings.
Plongeur bu xóot
Vocoder mooy jéego bu mujj bi ci gasoduc TTS yu bari: model bu melni Tacotron wala FastSpeech dafay wax luy waaja am ci spectrogram mel (nataalu fréquence bu dëgër ci diir bi), ba noppi vocoder bi dafay feesal misaali forme onde yi. Vocoder neural yu njëkk ya melni WaveNet dañu neexoon lool waaye dañu defar audio sample-par-sample, loolu tax ñu yeex lool. HiFi-GAN, bi Kong, Kim, ak Bae genne ci 2020, moo wecci bouclage autorégressif bi ak benn generatër buy joxe feed-forward buñ tàggat ci noon. Li gëna am solo mooy jëfandikoo ay diskriminatër yu bari yuy àtte audio bi ci eskaal yu wuute ak ci motif periodik yu wuute, di forse generatër bi mu am texture bu rafet bi ak periodicite pitch bi ci yoon. Resultaa bi mooy 22 kHz wax buñ synthesize téemeeri yoon gëna gaaw ci GPU, ak kalite buy xëcc dëgg-dëgg audio.
Gis-gis xarala
Generatëru HiFi-GAN dafay jël misaalu spectrogram mel bi jaaraleko ci ay convolusioŋ yuñ toxal, ak ay bloku champ yu bari yu ñuy jaxase kernel yu bari ak dilatation ngir jàpp motif vague yu bari. Ñaari famiy diskriminatër ñooy def poliis bi: benn diskriminatër bu bari-period dafay soppi siñaal 1D ci griy 2D ci prime yu melni 2, 3, 5, 7, 11 ngir jàpp periodicité pitch, ak benn diskriminatër bu bari-scale di xool forme vague ci yenn resolutions yu wàññeeku. Mel-spectrogram ak perte yu méngoo ak màndarga yi dañuy tax tàggat yaram bi nekk ci jàmm.
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 HiFi-GAN ak GAN
GAN vocoders ñu ngi gëna ndaw ak gaaw: doomi BigVGAN yokk ay aktivasioŋ anti-aliased ngir jëmmal ci waykat yi, jumtukaay yi, ak làkk yi ñu gisul, ci noonu la UnivNet ak Vocos di push ci universel, synthese all-band. Streaming ak aparey yi leegi dañuy doxal vocoding ci biir telefon yi ak earbuds yi ngir assistant yu néew latency. Lu gëna bari, diffusion ak xeetu audio yu méngoo ak debit ñu ngi leen di distiye ci generatër yu am benn pas yu nuroo ak GAN, di boole fidélité diffusion ak gaawaayu GAN. Xaarandil ni vocoder yi dina ñu dem ba nekk codec audio neural yu bari yuy dooleel wax ak music yépp.
Doxal ci àdduna dëgg
Defar kàdduy assistant virtuel yi ak aplikaasioŋu navigation yi soxla tontu te duñu yeexal dégg.
Dooleel jumtukaayi klonaasu baat ak dubbing ci jamono dëgg, fu ñuy soppi mel-spectrogram buñ klone ci audio buy sone ci naturel.
Dawal téere audiobook ak platform nettali podcast yuy boole waxtuy wax ci anam wu gaaw te yomb.
Dafay nekk seen xeetu vague ci biir sintetiseer yu way-baat ak demo music jaaraleko ci vocoder universel yu nuroo ak BigVGAN.
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 HiFi-GAN and GAN Vocoders 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
Vokoder WaveGAN paralel
Laaj yi ñuy faral di laaj
What is HiFi-GAN and GAN Vocoders?
HiFi-GAN vocoder buy génne ay noon la, muy soppi mel-spectrogram ci forme onde audio bu ñor ci saasi, di defar kàddu yu am kalite studio bu gëna gaaw ci jamono dëgg. Nekk na etape bu mujj biñ miin ci enregistrement text-to-speech ndax dafa gaaw, woyof, te jafe wuutale ko ak enregistrement dëgg.
Lan mooy liggéey bi njëkk ci vocoder bu melni HiFi-GAN ci pipeline bind-ci-kaddu?
Vocoder mooy etap bi mujj biy boole misaali forme onde yi ci mel-spectrogram bi model akustik defar.
Lu tax HiFi-GAN di gëna gaaw vocoder WaveNet bi njëkk?
WaveNet dafa defar misaalu audio ci misaal (autoregressivement), ci noonu la generatëru feed-forward bu HiFi-GAN di génne forme onde bi ci benn yoon, muy am gaawaay bu gëna gaaw ci jamono dëgg.
Lan mooy Discriminateur Multi-Period bu HiFi-GAN di jàppale ci jàpp?
Diskriminatëru diir yu bari yi dafay soppi jëmmu vague 1D ci 2D ci jëfandikoo diir yu am nimero bu njëkk ngir gis structure periodik biñ boole ak pitch bi.
GAN vocoders ñu ngi leen di tàggat 'ci noon.' Lu loolu di tekki fii?
Ci tabb GAN, generatër bi dafay jàng defar forme vague yu dëggu ba mëna nax reso discriminateur yiñ tàggat ngir ràññee dëgg ak audio synthetik.
Ban vocoder ci ginaaw mooy yokk HiFi-GAN ngir gëna mëna jëfandikoo kàddu yiñ gisul, way ak jumtukaay?
BigVGAN dafay yokk HiFi-GAN ba noppi yokk ay aktivaasioŋ yu bari yuy tere aliased, gëna yombal généralisasioŋ ci audio yi nekkul ci distribution yu melni way ak jumtukaay.