I-RHVoice 1.6.0, inkqubo yokudibanisa intetho evulelekileyo, ikhutshiwe. Ekuqaleni yaphuhliswa ukubonelela ngenkxaso esemgangathweni yesiRashiya, kwaye emva koko yalungiselelwa ezinye iilwimi, kuquka isiNgesi, isiPhuthukezi, isiUkraine, isiKyrgyz, isiTatar, kunye nesiGeorgia. Le khowudi ibhalwe nge-C++ kwaye isasazwa phantsi kwelayisenisi ye-LGPL 2.1. Inkxaso ye-GNU/Linux nayo iyaxhaswa.Linux, Windows и AndroidLe nkqubo iyahambelana neendlela eziqhelekileyo zokuguqula umbhalo ube yintetho: SAPI5 (Windows), Umsasazi weeNtetho (GNU/Linux) kunye Android I-API yoMbhalo-ukuya-kwiNtetho, kodwa ingasetyenziswa ne-NVDA screen reader. Umdali kunye nomphuhlisi oyintloko we-RHVoice ngu-Olga Yakovleva, oqhubeka nokuphuhlisa le projekthi nangona eyimfama ngokupheleleyo.
Inguqulelo entsha yongeza iindlela ezi-5 ezintsha zelizwi kwintetho yesiRashiya. Inkxaso yolwimi lwesiAlbania isetyenzisiwe. Isichazi-magama sesiUkraine sihlaziyiwe. Inkxaso yokubiza i-emoji yandisiwe. Ukulungiswa kweempazamo kusetyenzisiwe kwi-app yeqonga. Android, ukungeniswa kwezichazi-magama zomsebenzisi kwenziwe lula, kwaye inkxaso yeqonga yongeziwe Android 11. Izicwangciso ezintsha kunye nokusebenza kongezwe kwisiseko senjini, kuquka i-g2p.case, i-word_break, kunye nenkxaso yezihluzi zokulinganisa.
Masikhumbule ukuba i-RHVoice isebenzisa uphuhliso lweprojekthi ye-HTS (i-HMM/DNN-based Speech Synthesis System) kunye ne-parametric synthesis method with statistical models (Statistical Parametric Synthesis esekelwe kwi-HMM - i-Hidden Markov Model). Inzuzo yemodeli yezibalo ziindleko eziphezulu eziphantsi kunye nokungafuneki kwamandla e-CPU. Yonke imisebenzi yenziwa ekuhlaleni kwisixokelelwano somsebenzisi. Amanqanaba amathathu omgangatho wentetho axhaswayo (umgangatho ophantsi, uphezulu ukusebenza kunye nelifutshane ixesha lokuphendula).
Icala elisezantsi lemodeli yamanani lumgangatho ophantsi wokubizwa, ongafikeleli kwinqanaba le-synthesizers ezivelisa intetho esekwe kwindibaniselwano yamaqhekeza entetho yendalo, kodwa nangona kunjalo isiphumo sifundeka kakuhle kwaye sifana nokusasaza okurekhodiweyo kwisandisi-lizwi. . Ukuthelekisa, iprojekthi ye-Silero, ebonelela nge-injini yokudibanisa intetho evulekileyo ngokusekelwe kubuchwepheshe bokufunda koomatshini kunye nesethi yeemodeli zolwimi lwesiRashiya, iphezulu kumgangatho we-RHVoice.
Kukho iinketho zezwi ezili-13 ezifumanekayo ngolwimi lwesiRashiya, kunye ne-5 yesiNgesi. Kwizicwangciso unokutshintsha isantya, isantya kunye nevolumu. Ilayibrari ye-Sonic ingasetyenziselwa ukutshintsha i-tempo. Kuyenzeka ukuba uchonge ngokuzenzekelayo kwaye utshintshe iilwimi ngokusekwe kuhlalutyo lombhalo wegalelo (umzekelo, kumagama kunye neengcaphuno zolunye ulwimi, imodeli yokwenziwa kolu lwimi ingasetyenziswa). Iiprofayili zelizwi ziyaxhaswa, zichaza indibaniselwano yamazwi eelwimi ezahlukeneyo.
umthombo: opennet.ru
