Google yakaburitsa iyo Lyra odhiyo codec yekufambisa yekutaura mune yakashata yekubatanidza mhando

Google yakaunza itsva audio codec, Lyra, yakagadziridzwa kuti iwane yakanyanya kunaka mhando kunyangwe uchishandisa inononoka kutaurirana nzira. Iyo Lyra yekumisikidza kodhi yakanyorwa muC ++ uye yakavhurwa pasi peiyo Apache 2.0 rezenisi, asi pakati pezvinotsamira pakushanda pane proprietary raibhurari libsparse_inference.so ine kernel kuita kwekuverenga kwemasvomhu. Zvinocherechedzwa kuti raibhurari yevaridzi ndeyenguva pfupi - mune ramangwana Google inovimbisa kugadzira yakavhurika kutsiva uye nekupa rutsigiro kumapuratifomu akasiyana.

Panyaya yemhando yedhata rezwi rinofambiswa nekumhanya kwakaderera, Lyra yakanyanya kukwirira kune echinyakare macodecs anoshandisa madhijitari masaini ekugadzirisa nzira. Kuti uwane kufambiswa kwezwi kwemhando yepamusoro mumamiriro ehuwandu hushoma hweruzivo rwakafambiswa, mukuwedzera kune dzakajairwa nzira dzekutsikirira odhiyo uye kutendeuka kwechiratidzo, Lyra anoshandisa modhi yekutaura yakavakirwa pamuchina wekudzidza sisitimu, iyo inokutendera iwe kuti udzokorore ruzivo rusipo rwakavakirwa pa. matauriro chaiwo. Mhando yakashandiswa kugadzira ruzha yakadzidziswa nezviuru zvemaawa zvekurekodha manzwi mumitauro inodarika makumi manomwe.

Google yakaburitsa iyo Lyra odhiyo codec yekufambisa yekutaura mune yakashata yekubatanidza mhando

Iyo codec inosanganisira encoder uye decoder. Iyo encoder's algorithm inowira pasi kuti ibvise izwi data paramita ega ega makumi mana mamilliseconds, ichiamanikidza, uye nekuaendesa kune anogamuchira panetiweki. Nzira yekukurukurirana ine spidhi ye 40 kilobits pasekondi inokwana kuendesa data. Iwo akabviswa odhiyo paramita anosanganisira logarithmic mel spectrograms ayo anofunga nezvesimba maitiro ekutaura muakasiyana ma frequency renji uye anogadzirirwa achifunga nezve modhi yekunzwa kwemunhu.

Google yakaburitsa iyo Lyra odhiyo codec yekufambisa yekutaura mune yakashata yekubatanidza mhando

Iyo decoder inoshandisa generative modhi iyo, zvichibva pane inofambiswa maodhiyo paramita, inodzoreredza chiratidzo chekutaura. Kuti kuderedze kuomarara kwekuverenga, modhi yakareruka yakavakirwa pane inodzokororwa neural network yakashandiswa, inova musiyano weiyo WaveRNN yekutaura synthesis modhi, iyo inoshandisa yakaderera sampling frequency, asi inoburitsa masaini akati wandei anoenderana mumhando dzakasiyana dze frequency. Iwo masaini masaini anobva aiswa pamusoro kuti abudise chiratidzo chimwe chekubuda chinoenderana neyakatsanangurwa sampling rate.

Yakasarudzika processor mirairo inowanikwa mu64-bit ARM processors inoshandiswawo kumhanyisa. Nekuda kweizvozvo, kunyangwe nekushandiswa kwemuchina kudzidza, iyo Lyra codec inogona kushandiswa chaiyo-nguva yekutaura encoding uye decoding pamidhi-renji smartphones, ichiratidza chiratidzo chekufambisa latency ye90 milliseconds.

Source: opennet.ru

Voeg