Google ya buga codec audio na Lyra don watsa magana cikin rashin ingancin haɗin gwiwa

Google ya gabatar da sabon codec na audio, Lyra, wanda aka inganta don cimma iyakar ingancin muryar koda lokacin amfani da tashoshi na sadarwa a hankali. An rubuta lambar aiwatar da Lyra a cikin C ++ kuma an buɗe ƙarƙashin lasisin Apache 2.0, amma a cikin abubuwan dogaro da ake buƙata don aiki akwai ɗakin karatu na mallakar mallaka libsparse_inference.so tare da aiwatar da kernel don lissafin lissafi. An lura cewa ɗakin karatu na mallakar ɗan lokaci ne - a nan gaba Google ya yi alkawarin haɓaka buɗaɗɗen maye da bayar da tallafi ga dandamali daban-daban.

Dangane da ingancin bayanan muryar da ake watsawa a ƙananan gudu, Lyra ta fi girma fiye da codecs na gargajiya waɗanda ke amfani da hanyoyin sarrafa siginar dijital. Don cimma babban ingancin watsa murya a cikin yanayi na taƙaitaccen adadin bayanan da aka watsa, ban da hanyoyin al'ada na matsawa na sauti da jujjuya siginar, Lyra yana amfani da ƙirar magana dangane da tsarin koyon injin, wanda ke ba ku damar sake ƙirƙirar bayanan da suka ɓace bisa ga halaye na magana na al'ada. An horar da samfurin da aka yi amfani da shi don samar da sauti ta amfani da sa'o'i dubu da yawa na rikodin murya a cikin fiye da harsuna 70.

Google ya buga codec audio na Lyra don watsa magana cikin rashin ingancin haɗin gwiwa

Codec ɗin ya haɗa da encoder da mai ƙididdigewa. Algorithm na encoder yana tafasa ƙasa don fitar da sigogin bayanan murya kowane miliyon 40, yana matsa su, da aika su ga mai karɓa akan hanyar sadarwa. Tashar sadarwa mai gudun kilobits 3 a cikin dakika daya ta wadatar don watsa bayanai. Siffofin sauti da aka fitar sun haɗa da na'urar duban ra'ayi na logarithmic mel waɗanda ke yin la'akari da halayen kuzarin magana a cikin jeri daban-daban kuma an shirya su cikin la'akari da ƙirar tsinkayen sauraron ɗan adam.

Google ya buga codec audio na Lyra don watsa magana cikin rashin ingancin haɗin gwiwa

Mai ƙididdigewa yana amfani da ƙirar ƙira wanda, dangane da sigogin sauti da aka watsa, yana sake ƙirƙirar siginar magana. Don rage sarƙaƙƙiyar ƙididdiga, an yi amfani da ƙirar nauyi mai nauyi dangane da hanyar sadarwa mai maimaitawa, wanda shine bambance-bambancen ƙirar ƙirar magana ta WaveRNN, wanda ke amfani da ƙananan mitar samfur, amma yana haifar da sigina da yawa a layi daya a cikin jeri daban-daban. Sakamakon sigina ana sama da shi don samar da siginar fitarwa guda ɗaya daidai da ƙayyadadden ƙimar samfur.

Ana kuma amfani da umarnin sarrafawa na musamman da ake samu a cikin na'urori masu sarrafa 64-bit ARM don haɓakawa. A sakamakon haka, duk da yin amfani da na'ura koyo, da Lyra codec za a iya amfani da ainihin lokacin magana encoding da dikodi a kan tsakiyar kewayon wayoyin hannu, nuna latency watsa sigina na 90 millise seconds.

source: budenet.ru

Add a comment