Kaddu yu Gaaw ak TTS budul Dellu ginaaw
FastSpeech dafay defar spectrogramu kàddu bu mat sëkk ci paralel, du benn kadre benn yoon, loolu mooy tax synthese bi gëna gaaw te gëna dëgër.
Résumé
It solved the slow, error-prone generation that plagued earlier autoregressive models like Tacotron.
Plongeur bu xóot
Modèle TTS neuronal yu njëkk ya lu melni Tacotron 2 dañuy autoregressive: dañuy wax luy waaja am ci kadre audio bu nekk ci bi njëkk, te dafay yeex te dafay gaawa tëb wala baamtu kàddu sudee bàyyi xel bi jaarul yoon. FastSpeech, bi Microsoft ak Daara bu kawe bu Zhejiang dugal ci 2019, dafay soppi lii ci wax luy am ci kaadar yépp benn yoon. Reseau feed-forward bu sukkandiko ci Transformer dafay jël fonem yi, di wax bu baax ba ñaata fan la fonem bu nekk wara yàgg ak regulatëru guddaay bi, ba noppi di yaatal toppalante bi ba ci limu kaadar yi war balaa muy defar spectrogram bi ci benn yoon. FastSpeech 2 dafa gëna suqali lii ndax dafa wax luy waaja am ci ton bi ak doole ji, ba noppi di tàggat diir bi ñuy wut ci alignment forcé ludul distiller leen ci xeetu jàngalekat bu yeex, loolu mooy joxe wax ju gëna natural te ñu mëna ko saytu.
Gis-gis xarala
Li gëna am solo mooy yamale guddaay bi. Ndax bind ak audio dañu wuute ci guddaay, FastSpeech dafay wax luy yàgg ci fonem bu nekk, ba noppi baamtu limu nëbbu ci fonem bi lu bari yoon ngir méngoo ak guddaayi spectrogram bi. Liggéeyukaay bu leer bii mooy wecci bàyyi xel bu yomba dagg. Defar kadre bu nekk ci paralel dafay tekki ni diiru inference bi mingi aju ci guddaayi frase bi, te dindi bouclage autorégressive bi dafay dindi njuumti yi ci skip ak baamtu baat yi.
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 kàddu yu gaaw ak TTS yu dul dellu ginaaw
Synthese bu amul autoregressif mooy leegi li ñuy jëfandikoo ngir defar TTS ndax dafa gaaw, dëgër, te mën nañu ko saytu. Sistem yiy ñëw dañuy push ci wàllu seytu prosody bu gëna baax, streaming bu gëna néew latency ngir aplikaasioŋ yi ci saasi, ak variant yu mujj ba ci njeexte yuy sànni spectrogram bi ci digg bi yépp. Modèle yu amul autoregressive yu sukkandiko ci diffusion ak flow itam ñu ngi gëna yokk, di boole parallelism FastSpeech ak kalite generatif bu gëna am doole, ci noonu la pitch bu leer ak diir bi ñuy doxalee ñu ngi wéy di am solo ngir produit yu mëna soppali, yuy fësal kàddu.
Doxal ci àdduna dëgg
Aplikaasioŋu navigasioŋ yi ci saasi dañuy defar ci saasi ay laaj yu am kàddu yu ñuy jëfandikoo ci anamu FastSpeech.
Sistem IVR yiy serwiis kiliyaan yi dañuy soppi mbind dinaamik ci kàddu ci eskaal te du am benn njuumte ci sànni baat yi.
Jàngalekatu ekraŋ yu yomb yi dañuy defar kàddu yu gaaw te wóor ngir dokimaa yu gudd ci aparey yu woyof.
Jumtukaayi ëmbiitu baat yi dañuy may defarkat yi ñu mëna tànn ton ak ni ñuy waxee ci saasi, ndax FastSpeech 2 dafay wax lu leer ci ton ak energie.
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 FastSpeech and Non-Autoregressive TTS 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
TTS mbonaat
Laaj yi ñuy faral di laaj
What is FastSpeech and Non-Autoregressive TTS?
FastSpeech dafay defar spectrogramu kàddu bu mat sëkk ci paralel, du benn kadre benn yoon, loolu mooy tax synthese bi gëna gaaw te gëna dëgër. Dafa saafara jafe-jafe yi, jafe-jafe yi daan faral di am ci model autoregressif yu melni Tacotron.
Luy 'budul dellu ginaaw' ci wàllu FastSpeech?
Non-autoregressive dafay tekki ni dañuy defar kadre yi ci paralel ci benn yoon, duñu conditionné benn-benn ci kadre yi njëkk.
Ban jafe-jafe ci xeetu dellu ginaaw bu melni Tacotron 2 la FastSpeech gëna wax ci?
Foofu autoregressive mën na nekk lu jaarul yoon, moo waral ñuy sànni wala ñu baamtu kàddu yi, te dekodage bi ci toppalante dafay yeex; FastSpeech dafay defar ñoom ñaar.
Lan mooy wareefu 'régulatëru guddaay' ci FastSpeech?
Regulatëru guddaay bi dafay yaatal màndarga fonem yi ci seen guddaay biñ séentu, mu méngale mbind mi gëna gàtt ak spectrogram bi gëna gudd.
Lan la FastSpeech 2 yokk ci FastSpeech bi njëkk?
FastSpeech 2 dafay wax luy waaja am ci ton bi ak doole ji, ba noppi di jëfandikoo ay diir yuñ forse ngir méngale ko ak jàngalekat bu yeex, gëna yombal nature ak doxal.
Lan moo waral diiru inference bu FastSpeech di yokk bu baax ak guddaayu frase bi?
Ndax defar yi dañu paralel, yokk yeneen kadre du yokk jéego yu toppalante ni ko dekodage autorégressive di defee.