EnCodec, sabon Meta codec audio

encodec

Encodec codec ne wanda ke yanke hukunci ta amfani da hanyar sadarwa na jijiyoyi tare da adadin matsawa kusan 10x

Kwanan nan, Meta (Tsohon Facebook) ya bayyana sabon codec na sauti mai suna EnCodec, que yana amfani da dabarun koyon inji don ƙara yawan matsawa ba tare da rasa inganci ba.

Sabuwar hanyar za ta iya damfara da rage sauti a cikin ainihin lokaci don cimma raguwar girman zamani na zamani. codec da za a iya amfani da duka streaming audio a ainihin lokacin amma don ɓoyewa don ajiya na gaba a cikin fayiloli.

A yau, muna ba da cikakken bayani game da ci gaban da Muhimman Bincike na AI (FAIR) ya samu a fagen matsananciyar murya mai ƙarfi ta AI. Ka yi tunanin sauraron saƙon mai jiwuwa na abokinka a cikin yanki mara ƙarancin haɗin kai kuma baya tsayawa ko faɗuwa. Bincikenmu ya nuna yadda za mu iya amfani da AI don taimaka mana cimma wannan.

A cikin Codec bayar da samfura biyu shirye don saukewa:

  1. Samfurin dalili wanda ke amfani da ƙimar samfurin 24 kHz, yana goyan bayan sauti na monophonic kawai, kuma an horar da shi akan nau'ikan bayanan mai jiwuwa (wanda ya dace da rikodin magana). Ana iya amfani da samfurin don tattara bayanan odiyo don watsawa a ƙimar bit na 1,5, 3, 6, 12 da 24 kbps.
  2. Samfurin da ba dalili ba wanda ke amfani da ƙimar samfurin 48kHz, yana goyan bayan sautin sitiriyo, kuma an horar da shi akan kiɗa kawai. Samfurin yana goyan bayan ƙimar bit na 3, 6, 12 da 24 kbps.

Ga kowane samfurin, an shirya ƙarin samfurin harshe, mece yana ba da damar haɓaka mai mahimmanci a cikin matsawa rabo (har zuwa 40%) ba tare da asarar inganci ba. Ba kamar ayyukan da suka gabata ba don amfani da dabarun koyon injin don matse sauti, Ana iya amfani da EnCodec ba kawai don marufi na magana ba, har ma don matsawa kiɗa tare da mitar samfur na 48 kHz, daidai da matakin CD mai jiwuwa.

A cewar masu haɓaka sabon codec, ta hanyar watsawa a cikin ɗan ƙaramin adadin 64 kbps idan aka kwatanta da tsarin MP3, sun sami damar haɓaka ma'aunin matsawa na sauti da kusan sau goma yayin da suke riƙe daidaitaccen matakin inganci (misali, lokacin amfani da MP3). yana buƙatar bandwidth na 64 kbps, don canja wurin tare da inganci iri ɗaya a cikin EnCodec, 6 kbps ya isa).

Ana iya yanke wannan bayanan ta amfani da hanyar sadarwa na jijiyoyi. Mun sami kusan adadin matsawa na 10x idan aka kwatanta da MP3 a 64kbps, ba tare da asarar inganci ba. Yayin da aka bincika waɗannan fasahohin a baya don magana, mu ne farkon wanda ya sa ya yi aiki don 48 kHz samfurin sitiriyo audio (watau ingancin CD), wanda shine ma'auni don rarraba kiɗa.

Gine-gine na codec An gina shi a kan hanyar sadarwa ta jijiyoyi tare da "canji" gine-gine kuma yana dogara ne akan shaidu hudu: encoder, quantizer, decoder and discriminator:

  • El encoder yana fitar da sigogi daga bayanan muryar kuma yana canza shi zuwa rafi mai fakiti a ƙaramin firam.
  • El ƙididdiga (RVQ, Residual Vector Quantizer) yana jujjuya rafin fitarwar encoder zuwa saitin fakiti, yana matsa bayanin dangane da ƙimar bit ɗin da aka zaɓa. Fitar da ma'aunin ƙididdiga shine matsi na wakilcin bayanan da suka dace don watsawa akan hanyar sadarwa ko adanawa zuwa faifai.
  • El decoder yana yanke wakilcin bayanan da aka matsa kuma yana sake gina asalin sautin sautin.
  • El mai nuna wariya yana inganta ingancin samfuran da aka samar (samfurin) yin la'akari da samfurin tsinkayen sauraron ɗan adam.

Ba tare da la'akari da ingancin matakin da bitrate ba, ƙirar da aka yi amfani da su don ɓoyewa da yankewa sun bambanta cikin daidaitattun buƙatun albarkatu (ƙididdigar da ake buƙata don aiki na ainihi ana yin su akan ainihin CPU guda ɗaya).

A ƙarshe, ga waɗanda kuke sha'awar, ya kamata ku sani cewa aiwatar da aikin EnCodec an rubuta shi cikin Python ta amfani da tsarin PyTorch kuma an ba shi lasisi ƙarƙashin lasisin CC BY-NC 4.0 (Creative Commons Attribution-NonCommercial) don amfanin kasuwanci. kawai.

Idan kuna sha'awar ƙarin koyo game da shi, kuna iya tuntuɓar cikakkun bayanai a mahada mai zuwa.


Bar tsokaci

Your email address ba za a buga. Bukata filayen suna alama da *

*

*

  1. Alhakin bayanai: AB Internet Networks 2008 SL
  2. Manufar bayanan: Sarrafa SPAM, sarrafa sharhi.
  3. Halacci: Yarda da yarda
  4. Sadarwar bayanan: Ba za a sanar da wasu bayanan ga wasu kamfanoni ba sai ta hanyar wajibcin doka.
  5. Ajiye bayanai: Bayanin yanar gizo wanda Occentus Networks (EU) suka dauki nauyi
  6. Hakkoki: A kowane lokaci zaka iyakance, dawo da share bayanan ka.