Google isizindalwazi esinezichasiselo semisindo exubile engasetshenziswa ezinhlelweni zokufunda zomshini ezisetshenziselwa ukuhlukanisa imisindo exubile engafanele ibe izingxenye zayo ezingazodwana. Imodeli yokufunda yomshini ojulile ojwayelekile (TDCN++) nayo ishicilelwe engasetshenziswa ku-Tensorflow ukuhlukanisa imisindo. Idatha elungiselelwe ngokusekelwe eqoqweni ΠΈ ilayisensi ngaphansi kwe-CC BY 4.0.
Iphrojekthi eyethulwe i-FUSS (Ukuhlukaniswa Kwamahhala Komsindo Wendawo Yonke) kuhloswe ngayo ukuxazulula inkinga yokuhlukanisa noma iyiphi inombolo yemisindo engafanele, imvelo yayo engaziwa kusengaphambili. Ezinye izinhlelo ezifanayo ngokuvamile zilinganiselwe emsebenzini wokuhlukanisa phakathi kwemisindo ethile, njengamaphimbo nokungewona amazwi, noma abantu abahlukene abakhulumayo.
I-database iqukethe izingxube ezingaba yizinkulungwane ezingama-20. Ikhithi futhi ihlanganisa izimpendulo zomfutho wegumbi ezibalwe kusengaphambili kusetshenziswa isifanisi segumbi esakhelwe ngokwezifiso esicabangela ukuboniswa kodonga, indawo yomthombo womsindo, nendawo yemakrofoni.
Source: opennet.ru