Vocoders
Indangururamajwi ni moderi ihinduranya acoustic igereranya, mubisanzwe mel-spectrogram, muburyo bwumvikana bwumvikana.
Incamake
It is the final stage that gives modern text-to-speech and voice cloning their natural, human sound.
Kwibira cyane
Imvugo gakondo synthèse yakoresheje ibimenyetso-bitunganyiriza amajwi byakunze kumvikana cyangwa robot. Indangururamajwi ziga kubaka amajwi mbisi zivuye muri spekrogramu mumahugurwa kumasaha yafashwe. WaveNet (DeepMind, 2016) yari intambwe, guhanura amajwi icyarimwe icyarimwe icyarimwe 16,000+ kumasegonda, bitanga imvugo karemano ariko buhoro buhoro. Nyuma yaho moderi yacuruzaga kuri autoregressive icyuho cyihuta: WaveGlow yakoresheje igisekuru gishingiye kumurongo, Parallel WaveGAN na MelGAN bakoresheje imiyoboro yabyara inyungu, kandi HiFi-GAN yabaye igipimo cyamamare mugutanga amajwi yizerwa 22kHz yihuta cyane kuruta igihe nyacyo. Uyu munsi, amajwi hafi buri gihe igice cya kabiri cyumuyoboro wibyiciro bibiri, uhujwe na moderi ya acoustic nka Tacotron 2 cyangwa FastSpeech itanga mel-spectrogram.
Ubushishozi
Mel-spectrogramu ita amakuru yicyiciro cyamajwi, ikomeza gusa uburyo imbaraga zikwirakwizwa mumirongo yumurongo mugihe. Akazi katoroshye akazi ni uguhimba ibintu byemewe, bihuza umurongo ufite ubunini buhuye nibyinjira. Ijwi rishingiye kuri GAN nka HiFi-GAN rikoresha ivangura ryinshi rigenzura ibimenyetso ku munzani n'ibihe bitandukanye, bigasunika generator kubyara ibintu bifatika bifatika nk'ibihuza hamwe n'inzibacyuho ikarishye.
Ingaruka z'Ingamba
Kugera no kugera
Itezimbere kugerwaho binyuze mu kwandukura, kuvuga, no guhuza amajwi.
Igiciro na bije
Amatsinda yibitangazamakuru arashobora kohereza amajwi yihuse hamwe na bije nto.
Umuvuduko n'igipimo
Sisitemu ireba abakiriya irashobora gutunganya imikoranire ivugwa murwego runini.
Kazoza ka Vocoders
Vocoders iragenda iba nto kandi byihuse kuburyo ishobora gukora kuri terefone n'ibikoresho byashizwemo bidafite igicu. Hariho kandi gusunika ku majwi yisi yose asanzwe avuga abavuga bose, ururimi, kuririmba, cyangwa amajwi atavuga atongeye kwitoza. Inzira ibangikanye igabanya amajwi mu buryo butaziguye muri sisitemu ya nyuma kugeza ku ndunduro hamwe na codec ya neural, bigahindura umurongo uri hagati ya acoustic na flake ya fonctionnement no kugabanya ibihangano byatangijwe no kunyura muri ecran ya interineti.
Gushyira mu bikorwa Isi
Gukora amajwi yanyuma avugwa mumyandiko-y-abafasha nkabasomyi ba ecran na porogaramu zo kugenda
Gukora amajwi-asanzwe yumvikana clon muma dubbing hamwe nibikoresho byo kuvuga amajwi
Kongera kubaka amajwi yo kuririmba muri muzika ya AI hamwe na software-yerekana amajwi
Guha imbaraga ibikoresho-byijwi bisohoka kubavuga ubwenge hamwe nibikoresho bigerwaho nta seriveri izenguruka
Ingaruka & Kurinda
Gukoresha nabi amajwi no kwigira ibyago byiyongera mugihe uruhushya rubuze.
Ukuri kurashobora kugabanuka hejuru yimvugo, imvugo, cyangwa urusaku rwibidukikije.
Amajwi yubukorikori arashobora kwibeshya kumvugo yukuri nta kirango gisobanutse.
Igishushanyo mbonera
Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.
Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.
Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.
Andika amajwi yubukorikori kandi ugumane inyandiko zerekana kubazwa.
Komeza Ubushakashatsi
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 Neural 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
Ubuyobozi bukurikira
HiFi-GAN na GAN Vocoders
Ibibazo bikunze kubazwa
What is Neural Vocoders?
Indangururamajwi ni moderi ihinduranya acoustic igereranya, mubisanzwe mel-spectrogram, muburyo bwumvikana bwumvikana. Nicyiciro cyanyuma gitanga inyandiko igezweho-ku-jwi nijwi rikoresha amajwi yabo asanzwe, yumuntu.
Nuwuhe murimo wibanze wijwi ryijwi?
Indangururamajwi ifata imiterere ya acoustic yoroheje, mubisanzwe mel-spectrogram, kandi igahuza amajwi nyayo yibyuma wumva.
Niyihe moderi ya 2016 niyambere yatumye amajwi ya neural yumvikana neza?
WaveNet, kuva muri DeepMind mu 2016, yakoze amajwi yerekana icyitegererezo kandi itanga imvugo karemano itangaje, itangiza ibihe byijwi ryijwi.
Kuki WaveNet yumwimerere yatinze kubyara amajwi?
WaveNet ni autoregressive: buri cyitegererezo cyamajwi giterwa nicyabanje, bityo kubyara ibihumbi icumi byintangarugero kumasegonda byari bihenze kubara.
Ni ayahe makuru mel-spectrogramu yata cyane, bigatuma akazi k'ijwi katoroshye?
Mel-spectrograms igumana ubunini (ingufu kuri bande yumurongo) ariko igabanuka ryicyiciro, bityo amajwi agomba kongera kubaka imiterere ihuza umurongo icyiciro cyacyo kikaba cyemewe.
Nigute amajwi ashingiye kuri GAN nka HiFi-GAN atezimbere realism?
HiFi-GAN ikoresha ivangura rigenzura umunzani n'ibihe bitandukanye, bigasunika generator kubyara ibintu bifatika kandi bihinduka.