DiffWave Diffusion Vocoder
DiffWave wani vocoder ne na tushen watsawa wanda ke haɗa sauti ta hanyar ƙididdige hayaniyar bazuwar cikin tsarin igiyar ruwa, mai sharadi akan mel-spectrogram.
Dubawa
It brought diffusion models to high-fidelity speech, rivaling GANs and WaveNet without adversarial training.
Zurfafa nutsewa
DiffWave, Kong et al. a cikin 2020, yana amfani da tsarin ƙima mai yuwuwa yaduwa zuwa ingantaccen sauti. Yayin horo a hankali yana ƙara hayaniyar Gaussian zuwa tsarin tsaftataccen igiyar ruwa akan matakai da yawa, sannan ya koyi hanyar sadarwa don tsinkaya da cire wannan amo a kowane mataki. A lokacin tsarawa yana farawa daga tsattsauran amo kuma yana aiwatar da tsarin baya, wanda aka tsara akan mel-spectrogram, don dawo da tsaftataccen magana. Kashin baya shine hanyar sadarwa mara-autoregressive, diated-convolution network mai kama da WaveNet amma yana tsinkayar amo maimakon samfura. DiffWave yayi daidai da ƙaƙƙarfan vocoders cikin inganci kuma yana da ƙarfi musamman, har ma yana samar da madaidaicin magana mara sharadi da tabbataccen sakamako a cikin masu magana. Babban ciniki-kashe shine saurin: samfurin butulci yana buƙatar ɗimbin matakai zuwa dubunnan matakai, kodayake jadawalin gaggawa ya yanke wannan zuwa kaɗan zuwa shida.
Fahimtar Fasaha
DiffWave yana koyon gradient na rarraba bayanai a fakaice ta horar da hanyar sadarwa don tsinkayar amo da aka ƙara a matakin watsawa bazuwar, ta amfani da maƙasudin L2 mai sauƙi. Samfurin yana jujjuya ƙayyadaddun tsarin amo, kuma adadin matakai yana cinikin ingancin sauri; masu binciken sun gano a hankali zaɓaɓɓun jadawali na matakai kusan shida sun adana mafi yawan aminci, suna mai da tsarin mataki dubu zuwa wani abu mafi kusa da aiki.
Dabarun Tasiri
Shiga ku isa
Yana inganta samun dama ta hanyar rubutu, ba da labari, da mu'amalar murya.
Kudin da kasafin kuɗi
Ƙungiyoyin kafofin watsa labaru na iya jigilar sauti mai gogewa cikin sauri tare da ƙaramin kasafin kuɗi.
Gudu da sikelin
Tsarin fuskantar abokin ciniki na iya aiwatar da hulɗar magana a mafi girman ma'auni.
Makomar DiffWave Diffusion Vocoder
DiffWave ya ƙaddamar da vocoders na watsawa da magada masu sauri kamar PriorGrad da FastDiff waɗanda ke ƙidayar matakin matakin. Filin yana haɗuwa akan distillation da dabarun ƙirar ƙirar ƙira waɗanda ke nufin ɗaukar samfuri mai yaduwa ta mataki ɗaya, rufe tazarar gudu tare da masu muryoyin GAN yayin kiyaye ingantaccen horo da ƙarfi. Yi tsammanin ra'ayoyin watsawa don yaduwa zuwa cikin kiɗa, kodecs na jijiyoyi, da tsarar sauti na duniya inda yanayin ke da mahimmanci.
Aiwatar da Gaskiyar Duniya
Babban amintaccen jijiya rubutu-zuwa-magana baya ƙare wanda ke guje wa horarwar GAN mara ƙarfi
Ƙirƙirar magana mara sharadi don haɓaka bayanai da binciken sauti
Haɗin murya mai ƙarfi-ƙarfi inda samfurin ɗaya ke sarrafa yawancin muryoyin akai-akai
Wurin gwaji don bincike-binciken yaɗuwa da sauri, yin amfani da gajeriyar jadawalin amo zuwa sauti na ainihi
Hatsari & Tsare-tsare
Rashin amfani da murya da haɗarin kwaikwaya yana ƙaruwa lokacin da aka rasa izini.
Daidaituwa na iya faɗuwa cikin lafuzza, yaruka, ko mahalli masu hayaniya.
Ana iya kuskuren sauti na roba don ingantacciyar magana ba tare da bayyananniyar lakabi ba.
Taswirar Hanya
Sami tabbataccen izini don ɗaukar murya, cloning, da sake amfani.
Gwajin ingantattun masu magana daban-daban da yanayin baya.
Ƙayyade lokacin da dole ne ɗan adam ya duba ko ya amince da abubuwan da aka fitar.
Yi lakabin sauti na roba da kuma adana bayanan da aka tabbatar don yin lissafi.
Ci gaba da Bincike
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 DiffWave Diffusion 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
Jagora na gaba
Stable Audio Latent Diffusion
Tambayoyin da ake yawan yi
What is DiffWave Diffusion Vocoder?
DiffWave wani vocoder ne na tushen watsawa wanda ke haɗa sauti ta hanyar ƙididdige hayaniyar bazuwar cikin tsarin igiyar ruwa, mai sharadi akan mel-spectrogram. Ya kawo samfuran watsawa zuwa babban magana mai aminci, gasa GANs da WaveNet ba tare da horo na gaba ba.
Ta yaya DiffWave ke samar da sauti a lokacin ƙididdigewa?
DiffWave yana farawa daga hayaniyar Gaussian kuma yana gudanar da tsarin jujjuyawa, akai-akai yana musantawa yayin da aka sanya shi akan mel-spectrogram, don samar da tsarin kalaman.
Menene ainihin hanyar sadarwar DiffWave ke koya don tsinkaya yayin horo?
DiffWave yana horar da hanyar sadarwa don hasashen hayaniyar Gaussian da aka ƙara a wani matakin watsawa da aka zaɓa, ta amfani da asarar L2 mai nauyi.
Menene babban koma baya na DiffWave idan aka kwatanta da GAN vocoders?
Samfurin yaɗuwar butulci yana buƙatar matakai da yawa na baya; kodayake jadawalai masu sauri suna taimakawa, tsarin maimaitawa shine babban farashin saurin gudu.
Wace fa'ida DiffWave ke da shi akan masu vocoders na GAN a horo?
An horar da samfuran watsawa tare da tabbataccen manufar salon koma-baya, tare da kawar da rashin zaman lafiyar da horon GAN na gaba zai iya wahala.
Wane irin gine-ginen cibiyar sadarwa-hasashen amo-Wave yayi kama?
DiffWave yana amfani da kashin baya mara-autoregressive na rikice-rikice masu kama da WaveNet, amma yana tsinkayar amo maimakon samfuran sauti.