Google yatulutsa mawu a Lyra audio codec kuti azilankhula mosalumikizana bwino

Google yakhazikitsa codec yatsopano yomvera, Lyra, yokonzedwa kuti ikwaniritse mawu abwino kwambiri ngakhale mutagwiritsa ntchito njira zolumikizirana pang'onopang'ono. Khodi yoyendetsera Lyra imalembedwa mu C ++ ndipo imatsegulidwa pansi pa layisensi ya Apache 2.0, koma pakati pa zodalira zomwe zimafunikira kuti zigwire ntchito pali laibulale ya eni libsparse_inference.so yokhala ndi kukhazikitsa kernel pakuwerengera masamu. Zimadziwika kuti laibulale ya eni ndi yanthawi yochepa - m'tsogolomu Google ikulonjeza kuti ipanga chosinthira chotseguka ndikupereka chithandizo pamapulatifomu osiyanasiyana.

Pankhani yamtundu wa data yotumiza mawu pa liwiro lotsika, Lyra ndi wapamwamba kwambiri kuposa ma codec achikhalidwe omwe amagwiritsa ntchito njira zosinthira ma digito. Kuti akwaniritse kufala kwa mawu apamwamba pazikhalidwe zochepa za chidziwitso chofalitsidwa, kuwonjezera pa njira zochiritsira zamawu ndi kutembenuka kwazizindikiro, Lyra amagwiritsa ntchito chilankhulo chochokera pamakina ophunzirira makina, omwe amakulolani kubwereza zomwe zikusowa potengera mawonekedwe amawu. Chitsanzo chomwe chinagwiritsidwa ntchito popanga mawuwo chinaphunzitsidwa pogwiritsa ntchito mawu ojambulitsa mawu a maola masauzande angapo m’zinenero zoposa 70.

Google yatulutsa mawu a Lyra audio codec kuti azilankhula mosalumikizana bwino

Codec imaphatikizapo encoder ndi decoder. Ma algorithm a encoder amafika potulutsa magawo a data ya mawu pa ma milliseconds 40 aliwonse, kuwapanikiza, ndikuwatumiza kwa wolandila pa netiweki. Njira yolumikizirana yomwe ili ndi liwiro la 3 kilobits pamphindikati ndiyokwanira kutumiza deta. Magawo amawu ochotsedwa amaphatikiza ma logarithmic mel spectrogram omwe amaganizira za mphamvu zamalankhulidwe m'magawo osiyanasiyana ndipo amakonzedwa potengera chitsanzo cha makutu a munthu.

Google yatulutsa mawu a Lyra audio codec kuti azilankhula mosalumikizana bwino

Decoder imagwiritsa ntchito njira yopangira yomwe, kutengera magawo amawu operekedwa, imapanganso chizindikiro chakulankhula. Pofuna kuchepetsa zovuta zowerengera, chitsanzo chopepuka chochokera ku neural network yobwerezabwereza chinagwiritsidwa ntchito, chomwe chiri chosiyana cha WaveRNN kulankhula kaphatikizidwe ka mawu, komwe kumagwiritsa ntchito maulendo otsika a sampuli, koma kumapanga zizindikiro zingapo zofanana mumayendedwe osiyanasiyana. Zizindikiro zotsatiridwazi zimayikidwa pamwamba kuti zipange chizindikiro chimodzi chofanana ndi mlingo womwe waperekedwa.

Malangizo apadera a purosesa omwe amapezeka mu 64-bit ARM processors amagwiritsidwanso ntchito kuti apititse patsogolo. Chotsatira chake, ngakhale kugwiritsa ntchito makina ophunzirira, codec ya Lyra ikhoza kugwiritsidwa ntchito polemba nthawi yeniyeni ndi kulembera ma foni apakati pa mafoni apakati, kusonyeza kuchedwa kwa mauthenga a 90 milliseconds.

Source: opennet.ru

Kuwonjezera ndemanga