OkulandelayoUmhlahlandlela olandelayo
Umculo we-AI
Umsindo we-AI
UMHLAHLANDLELA WE-AI womsindo
Isandiso somculo we-AI nokuqhubeka
Isandiso somculo we-AI sisebenzisa imodeli ekhiqizayo ukuze siqhubeke nesiqeshana somsindo esikhona, sikhiqize kabusha isigaba esimaphakathi (ukupenda), senze isitayela kabusha, noma sengeze futhi sishintshe iziqu zensimbi ngayinye.
Kuleli khasi4 amaminithi afundiwe
Uhlolojikelele
Ivumela umbono omfushane ukhule ube yingoma egcwele noma ukuthatha okunephutha kulungiswe ngaphandle kokurekhoda kabusha. Ibalulekile ngoba ishintsha indlela abantu abawubhala futhi bahlele ngayo umculo, kodwa inemikhawulo yekhwalithi ecacile futhi ayisusi amalungelo anoma ubani ongumnikazi womsindo wokuqala.
I-Deep Dive
Isandiso sithatha isiqeshana esikhona futhi sikhiqize okulandelayo. Enye indlela evamile ukuqhubeka kwe-autoregressive. I-MusicGen ye-Meta, ekhishwe ngo-2023 njengengxenye ye-AudioCraft, iguqula umsindo ube amathokheni ahlukene nge-neural codec ebizwa ngokuthi i-EnCodec futhi ibikezela amanye amathokheni ngemva kwalawo owanikezayo, njengoba nje imodeli yolimi iqhubeka nomusho. Izinsizakalo zabathengi ezifana ne-Suno ne-Udio zinikeza lokhu njengesici esithi Nweba, ngokuvamile esikuvumela ukuthi ukhethe indawo yokuqala futhi ushintshe amagama wezinhlamvu noma isitayela sesigaba esisha. Ukupenda kukhiqiza kabusha indawo efihliwe phakathi kuyilapho kugcinwa umsindo nhlangothi zombili, kuwusizo ekulungiseni ingoma noma ukushintsha amabha amane. Amamodeli okusabalalisa afaneleka lokhu ngoba umsindo owaziwayo ungabanjwa ungaguquki kuyilapho kukhishwa isikhala esifihliwe kuphela. Amamodi omsindo ukuya komsindo noma we-remix, njengaleyo i-Stability AI eyethulwe nge-Stable Audio 2.0 ngo-2024, ingeza kancane umsindo kokokufakayo futhi iyenze kabusha ngaphansi kokwaziswa okusha, ukuze isakhiwo siphile ngenkathi kushintsha i-timbre nesitayela. Isilungiselelo samandla silawula ukuthi kungakanani okwangempela okusele. Umsebenzi we-stem uhlanganisa ukuhlukanisa nokukhiqiza. Amamodeli okuhlukanisa imithombo afana ne-Meta's Open-source Demucs ahlukanisa ingxube ibe amazwi, izigubhu, amabhesi nokunye; ijeneretha ingase ingeze noma ishintshe ingxenye eyodwa enesimo kokunye. Imikhawulo ingokoqobo. Izandiso ezinde ziyakhukhuleka: i-tempo, ukhiye, ibhalansi yokuxuba kanye nezwi lezwi lingazulazula. Ukujoyina kungachofoza noma kugxume ngokuzwakalayo. Amamodeli abona iwindi lokuqukethwe okulinganiselwe, ngakho-ke i-motif evela kumzuzu wokuqala ingase ilibaleke. Umbono oyiphutha ovamile ukuthi umfuziselo uqonda ukwakheka kwengoma ngendlela umculi enza ngayo; ihambisana namaphethini kumsindo wakamuva. Amalungelo ayaphikelela: ukwelula ukurekhodwa kwezentengiso okungekona okwakho kudala okuphuma kukho, futhi imigomo yezinsizakalo eziningi iyakwenqabela ukulayisha umsindo ongenawo amalungelo kuwo.
I-Strategic Impact
Finyelela futhi ufinyelele
Ithuthukisa ukufinyeleleka ngokuloba, ukulandisa, nezixhumi ezibonakalayo zezwi.
Izindleko kanye nesabelomali
Amaqembu emidiya angathumela umsindo opholishiwe ngokushesha ngamabhajethi amancane.
Isivinini nesikali
Amasistimu abhekene nekhasimende angacubungula ukusebenzelana okukhulunyiwe ngesilinganiso esikhulu.
Ikusasa le-AI Music Extension and Continuation
Ucwaningo luphokophela kumongo omude, ukuqwashisa ngesakhiwo okungcono, kanye nokulawula okucolisekileyo phezu kweziqu ngazinye, okufanele kunciphise ukukhukhuleka futhi kwenze ukuhlelwa kwezinga lesigaba kubikezelwe. Ukuhlanganisa kuziteshi zokusebenza zomsindo wedijithali kuwumhlahlandlela okungenzeka, njengoba ukuhlela ngaphakathi kwephrojekthi ekhona kusiza kakhulu kunokukhiqiza izingoma eziphelele esipheqululini. Izinga eliphezulu kanye nokulondolozwa kobunikazi bomculi othile kuseyizinkinga ezinzima. Ngasohlangothini lwamalungelo, ukuhlolwa kokulayisha nokuhlelwa kwelayisensi phakathi kwamasevisi nabaphathi bamalungelo kusathuthuka, ngakho lokho abasebenzisi abangakwelula ngokusemthethweni kuzoncika emigomeni eguqukayo kanye nemiphumela yenkantolo esikhundleni sobuchwepheshe kuphela.
Ukuqaliswa Komhlaba Wangempela
Umbhali wengoma uhosha umbono wekhorasi wamasekhondi angu-20, awulayishe kujeneretha womculo, futhi asebenzise isici esithi Nweba sisuka kumaki angu-0:18 ukuze akhiqize ivesi nebhuloho elilandela ngokhiye ofanayo kanye ne-tempo.
Umhleli we-podcast unombhede womculo onelayisense wamasekhondi angu-30 omfushane kakhulu esingeniso, ngakho wenza ukuqhubeka futhi awuhlukanise emugqeni webha ukuze afinyelele imizuzwana engama-60.
Umdidiyeli uhlukanisa idemo endala ibe amazwi, izigubhu, ibhesi nezinye iziqu nge-Demucs, uthulisa isiqu sesigubhu esibuthakathaka, futhi unemodeli ekhiqiza ingxenye yesigubhu entsha ebekwe eziqwini ezisele.
Ibhendi ipenda amabha amabili lapho umculi eguqule ilirikhi, egcina umsindo nhlangothi zombili, bese ihlola ukuthi izwi elivuselelwe kabusha lisazwakala njengomculi ofanayo.
Izingozi & Guardrails
Ukusetshenziswa kabi kwezwi kanye nezingozi zokuzenza ongeyena ziyanda uma imvume ingekho.
Ukunemba kungase kwehle kuzo zonke izinhlobo zokuphimisela, izilimi zesigodi, noma izindawo ezinomsindo.
Umsindo wokwenziwa ungenziwa iphutha njengenkulumo eyiqiniso ngaphandle kokulebula okucacile.
Ukuqalisa Umhlahlandlela
Thola imvume esobala yokuthwebula izwi, ukuhlanganisa, nokusebenzisa kabusha.
Ikhwalithi yokuhlola kuzo zonke izipikha nezimo zangemuva.
Chaza ukuthi kunini lapho umuntu kufanele abuyekeze noma agunyaze okuphumayo.
Lebula umsindo wokwenziwa futhi ugcine amarekhodi atholakalayo ukuze aziphendulele.
Qhubeka Uhlole
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 AI Music Extension and Continuation 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
Imibuzo evame ukubuzwa
Kuyini Isandiso Somculo we-AI nokuqhubeka?
Isandiso somculo we-AI sisebenzisa imodeli ekhiqizayo ukuze siqhubeke nesiqeshana somsindo esikhona, sikhiqize kabusha isigaba esimaphakathi (ukupenda), senze isitayela kabusha, noma sengeze futhi sishintshe iziqu zensimbi ngayinye. Ivumela umbono omfushane ukhule ube yingoma egcwele noma ukuthatha okunephutha kulungiswe ngaphandle kokurekhoda kabusha. Ibalulekile ngoba ishintsha indlela abantu abawubhala futhi bahlele ngayo umculo, kodwa inemikhawulo yekhwalithi ecacile futhi ayisusi amalungelo anoma ubani ongumnikazi womsindo wokuqala.
Imodeli ye-autoregressive njenge-MusicGen iqhubeka kanjani nesiqeshana somsindo?
Isiqeshana siguqulwa sibe amathokheni ahlukene futhi imodeli ibikezela ukuthi yimaphi amathokheni alandelayo, okufana nendlela imodeli yolimi eqhubeka ngayo nombhalo.
Iyiphi i-neural codec esetshenziswa i-MusicGen ukuze iguqule umsindo ube amathokheni?
I-MusicGen ithembele ku-EnCodec, i-neural codec yomsindo evela ku-Meta, ukuze imele umsindo njengamathokheni ahlukene.
Kwenzani ukupenda okulalelwayo?
I-inpainting igcwalisa noma imiselela i-span ekhethiwe, efana ne-lyric eguquguqukayo, kuyilapho igcina umsindo ozungezile unjalo.
Kungani amamodeli okusabalalisa ekufanelekela kahle ukupenda?
Ukusabalalisa kungacindezela izifunda ezaziwayo phakathi nokukhipha umsindo, ngakho-ke isigaba esilahlekile kuphela esivuselelwa kabusha ukuze silingane nendawo yaso.
Kumodi ye-remix noma yomsindo-kuya-umsindo, yini isilungiselelo samandla esilawulayo?
Amandla aphakeme angeza umsindo owengeziwe futhi avuselele ngokwengeziwe, ashintshe okokufaka ngokuqhubekayo; amandla aphansi agcina isakhiwo sawo esiningi.
Qhubeka ufunda
Iziqondiso ezihlobene
Imihlahlandlela eyengeziwe yalesi sihloko