Kompania Google prezantoi kodekun e ri audio Lyra, i optimizuar p\u00ebr t\u00eb arritur cil\u00ebsin\u00eb maksimale t\u00eb transmetimit t\u00eb z\u00ebrit edhe gjat\u00eb p\u00ebrdorimit t\u00eb kanaleve shum\u00eb t\u00eb ngadalta. Kodi i implementimit t\u00eb Lyra \u00ebsht\u00eb shkruar n\u00eb C++ dhe \u00ebsht\u00eb i hapur n\u00ebn licenc\u00ebn Apache 2.0, por n\u00eb mesin e var\u00ebsive t\u00eb nevojshme p\u00ebr funksionimin e tij p\u00ebrfshihet biblioteka pron\u00ebsore libsparse_inference.so me implementimin e b\u00ebrtham\u00ebs p\u00ebr llogaritje matematikore. V\u00ebrehet se biblioteka pron\u00ebsore \u00ebsht\u00eb p\u00ebrkoh\u00ebsore \u2014 m\u00eb von\u00eb Google premton se do t\u00eb zhvilloj\u00eb nj\u00eb z\u00ebvend\u00ebsim t\u00eb hapur dhe do t\u00eb siguroj\u00eb mb\u00ebshtetje p\u00ebr platforma t\u00eb ndryshme. <\/p>\n
N\u00eb cil\u00ebsin\u00eb e t\u00eb dh\u00ebnave t\u00eb z\u00ebrit t\u00eb transmetuara n\u00eb shpejt\u00ebsi t\u00eb ulta, Lyra tejkalon ndjesh\u00ebm kodek\u00ebt tradicional\u00eb, t\u00eb cil\u00ebt p\u00ebrdorin metoda t\u00eb p\u00ebrpunimit t\u00eb sinjaleve digjitale. P\u00ebr t\u00eb arritur cil\u00ebsi t\u00eb lart\u00eb t\u00eb transmetimit t\u00eb z\u00ebrit n\u00eb kushte t\u00eb kufizuara t\u00eb informacionit t\u00eb transmetuar, p\u00ebrve\u00e7 metodave standarde t\u00eb kompresimit t\u00eb z\u00ebrit dhe transformimit t\u00eb sinjaleve, n\u00eb Lyra aplikohet nj\u00eb model z\u00ebri mbi baz\u00ebn e nj\u00eb sistemi t\u00eb m\u00ebsimit t\u00eb makinerive, q\u00eb lejon rimarrjen e informacionit t\u00eb munguar mbi baz\u00ebn e karakteristikave tipike t\u00eb z\u00ebrit. Modeli i angazhuar p\u00ebr gjenerimin e z\u00ebrit \u00ebsht\u00eb m\u00ebsuar duke p\u00ebrdorur disa mij\u00ebra or\u00eb regjistrimesh t\u00eb z\u00ebrit n\u00eb m\u00eb shum\u00eb se 70 gjuh\u00eb.
<\/center> <\/p>\n
Kodeku p\u00ebrfshin nj\u00eb kodues dhe nj\u00eb dekodues. Algoritmi i funksionimit t\u00eb koduesit konsiston n\u00eb nxjerrjen e parametrave t\u00eb t\u00eb dh\u00ebnave t\u00eb z\u00ebrit \u00e7do 40 milisekonda, kompresimin e tyre dhe d\u00ebrgimin e tyre tek marr\u00ebsi p\u00ebrmes rrjetit. P\u00ebr t\u00eb transmetuar t\u00eb dh\u00ebnat, mjafton nj\u00eb kanal komunikimi me shpejt\u00ebsi 3 kilobit n\u00eb sekond\u00eb. Parametrat e z\u00ebrit t\u00eb nxjerra p\u00ebrfshijn\u00eb mel-spektrogramet logaritmike, t\u00eb cilat marrin parasysh karakteristikat e energjis\u00eb s\u00eb z\u00ebrit n\u00eb frekuenca t\u00eb ndryshme dhe jan\u00eb p\u00ebrgatitur duke marr\u00eb n\u00eb konsiderat\u00eb modelin e perceptimit njer\u00ebzor t\u00eb d\u00ebgjimit.
<\/center> <\/p>\n
Dekoderi p\u00ebrdor nj\u00eb model gjenerativ, i cili, bazuar n\u00eb parametrat e dh\u00ebn\u00eb t\u00eb z\u00ebrit, ri-krijon sinjalin me z\u00eb. P\u00ebr t\u00eb reduktuar kompleksitetin e llogaritjeve, \u00ebsht\u00eb aplikuar nj\u00eb model i leht\u00eb i bazuar n\u00eb nj\u00eb rrjet nervor t\u00eb rikthyer, i cili \u00ebsht\u00eb nj\u00eb variant i modelit t\u00eb sintez\u00ebs s\u00eb z\u00ebrit WaveRNN, ku p\u00ebrdoret nj\u00eb frekuenc\u00eb m\u00eb e ul\u00ebt e mostrave, por gjenerohen paralelisht disa sinjale n\u00eb frekuenca t\u00eb ndryshme. Sinjalet e marra m\u00eb pas p\u00ebrfshihen p\u00ebr t\u00eb marr\u00eb nj\u00eb sinjal t\u00eb vet\u00ebm dalje, q\u00eb korrespondon me frekuenc\u00ebn e k\u00ebrkuar t\u00eb diskriminimit. <\/p>\n
P\u00ebr t\u00eb shpejtuar gjithashtu jan\u00eb p\u00ebrdorur instruksione t\u00eb specializuara p\u00ebr procesorin, t\u00eb disponueshme n\u00eb procesor\u00ebt 64-bit ARM. Si rezultat, pavar\u00ebsisht nga p\u00ebrdorimi i m\u00ebsimit makinerik, kodeku Lyra mund t\u00eb p\u00ebrdoret p\u00ebr kodimin dhe dekodimin e z\u00ebrit n\u00eb koh\u00eb reale n\u00eb smartphone me \u00e7mime mesatare, duke demonstruar nj\u00eb vones\u00eb t\u00eb transmetimit t\u00eb sinjalit n\u00eb nivelin e 90 milisekondave. \n Burimi: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"
![\"Google](\"\/wp-content\/uploads\/2021\/04\/dff1cbc3a2422197df2b4510b0a4fc75.png\")
<\/center> <\/p>\n![\"Google](\"\/wp-content\/uploads\/2021\/04\/5fedbeb0d9e364c54c9fd35dab76f019.png\")
<\/center> <\/p>\n