Facebook inoburitsa EnCodec odhiyo codec vachishandisa muchina kudzidza

Meta/Facebook (yakarambidzwa muRussian Federation) yakaunza itsva odhiyo codec, EnCodec, iyo inoshandisa muchina nzira dzekudzidza kuwedzera chiyero chekumanikidza pasina kurasikirwa nemhando. Iyo codec inogona kushandiswa ese ari maviri ekutepfenyura odhiyo munguva chaiyo uye encoding yekuzochengeta gare gare mumafaira. Iyo EnCodec referensi yekushandisa yakanyorwa muPython uchishandisa iyo PyTorch chimiro uye ine rezinesi pasi peCC BY-NC 4.0 (Creative Commons Attribution-NonCommercial) rezinesi rekushandiswa kusiri kwekutengesa chete.

Mhando mbiri dzakagadzirirwa-dzakagadzirwa dzinopihwa kurodha:

  • Iyo causal modhi inoshandisa 24 kHz sampling rate, inotsigira chete monophonic odhiyo, uye yakadzidziswa pane akasiyana odhiyo data (inokodzera kutaura coding). Iyo modhi inogona kushandiswa kurongedza odhiyo data yekufambisa pabit rates ye1.5, 3, 6, 12 uye 24 kbps.
  • Iyo isiri-causal modhi inoshandisa sampling rate ye48 kHz, inotsigira stereo odhiyo uye yakadzidziswa chete pamimhanzi. Iyo modhi inotsigira bitrate ye3, 6, 12 uye 24 kbps.

Pamuenzaniso wega wega, imwe yemhando yemutauro wakagadzirirwa, iyo inokubvumira kuti uwane kuwedzera kukuru kwehuwandu hwekugadzirisa (kusvika ku40%) pasina kurasikirwa kwehutano. Kusiyana nemapurojekiti akambogadzirwa achishandisa nzira dzekudzidza muchina wekumanikidza odhiyo, EnCodec inogona kushandiswa kwete kungorongedza yekutaura, asiwo yekumanikidza mimhanzi ine sampling rate ye48 kHz, inoenderana nechiyero cheaudio CD. Sekureva kwevagadziri veiyo codec nyowani, pakufambisa neiyo bitrate ye64 kbps zvichienzaniswa neiyo MP3 fomati, vakakwanisa kuwedzera dhigirii redhidhiyo kanenge kagumi vachichengeta chiyero chakafanana chemhando (semuenzaniso, pakushandisa. MP3, bandwidth ye64 kbps inodiwa, kuitira kutapurirana neiyo mhando imwechete muEnCodec inokwana 6 kbps).

Iyo codec architecture yakavakirwa pane neural network ine "transformer" architecture uye yakavakirwa pane ina link: encoder, quantizer, decoder uye kusarura. Iyo encoder inobvisa iyo paramita yezwi data uye inoshandura rwizi rwakazara kune yakaderera furemu chiyero. Iyo quantizer (RVQ, Residual Vector Quantizer) inoshandura kubuda kwerukova ne encoder kuita seti yemapakiti, kudzvanya ruzivo zvichibva pane yakasarudzwa bitrate. Iyo yakabuda ye quantizer inomiririra yakamanikidzwa yedata, yakakodzera kutapurirana pamusoro petiweki kana kuchengetedza kune dhisiki.

Iyo decoder inosarudza iyo yakamanikidzwa inomiririra yedata uye inovakazve yekutanga ruzha wave. Musarura anovandudza hutano hwemasampuli akagadzirwa, achifunga nezvemuenzaniso wekunzwa kwevanhu. Pasinei nechiyero chemhando uye bitrate, mamodheru anoshandiswa encoding uye decoding anosiyaniswa neane mwero zviwanikwa zvekushandisa (maverengero anodiwa pakushanda-nguva chaiyo anoitwa pane imwechete CPU musimboti).

Facebook inoburitsa EnCodec odhiyo codec vachishandisa muchina kudzidza


Source: opennet.ru

Voeg