uphando ugcino lwedatha yereferensi yezandi ezixubeneyo ezinokusetyenziswa kwiinkqubo zokufunda zoomatshini ezisetyenziselwa ukwahlula izandi ezixubeneyo ezingaqondakaliyo zibe ngamalungu azo. Imodeli yokufunda koomatshini obunzulu begeneric (TDCN++) nayo iye yapapashwa enokusetyenziswa kwi-Tensorflow ukwahlula izandi. Idatha ilungiswe ngokusekelwe kwingqokelela ΠΈ inikwe ilayisenisi phantsi kweCC BY 4.0.
Iprojekthi enikezelweyo yeFUSS (iFree Universal Sound Separation) ijolise ekusombululeni ingxaki yokwahlula naliphi na inani lezandi ezingaqondakaliyo, uhlobo lwalo olungaziwa kwangaphambili. Ezinye iisistim ezifanayo ngokubanzi zilinganiselwe kumsebenzi wokwahlula phakathi kwezandi ezithile, njengamazwi kunye nabangengawo amazwi, okanye abantu abahlukeneyo abathethayo.
I-database iqulethe malunga nama-20 amawaka okuxuba. Ikiti iphinda ibandakanye iimpendulo zempembelelo yegumbi elibalwe kwangaphambili kusetyenziswa i-simulator yegumbi elakhiwe ngokwezifiso ezithatha ukubonakalisa udonga, indawo yomthombo wesandi, kunye nendawo yemakrofoni.
umthombo: opennet.ru