EnCodec Audio funmorawon
EnCodec jẹ Meta's kodẹki ohun iṣan ti o ni ifaramọ ti o ga julọ ti o rọ ọrọ ati orin pọ ni awọn iwọn biiti kekere pupọ pẹlu idije didara ti awọn ọna kika ti o wuwo pupọ julọ.
Akopọ
It matters because it underpins modern generative audio systems and ships in open-source form for anyone to use.
Jin Dive
Ti tu silẹ nipasẹ Meta AI ni ọdun 2022, EnCodec tẹle SoundStream blueprint ti encoder kan, kuantizer vector ti o ku (RVQ), ati decoder ti o kọ ẹkọ ipari si ipari, ṣugbọn o ṣe afikun awọn atunṣe pupọ. O nlo kooduopo convolutional ti o lagbara-sisanwọle, spectrogram-ọpọlọpọ ati awọn adanu atunkọ akoko-akoko, ati awọn alayatọ ọta fun didara oye. Ilowosi pataki kan jẹ awoṣe entropy ti o da lori Transformer kekere ti o tun pọ si awọn koodu ti o ni iwọn laisi ipadanu, fifin awọn ege afikun jade laisi pipadanu didara. EnCodec tun ṣafihan iwọntunwọnsi ti o ṣe iwọn laifọwọyi awọn adanu ikẹkọ idije pupọ ki wọn duro iduroṣinṣin. O mu monophonic 24 kHz ati ohun afetigbọ sitẹrio 48 kHz, nṣiṣẹ kọja awọn iwọn biiti bi 1.5, 3, 6, ati 12 kbps, ati ni 6 kbps de didara ti o ni afiwe si MP3 ni 64 kbps. Agbara awọn ami rẹ Meta's MusicGen ati AudioGen.
Imọ-imọ-ẹrọ
EnCodec's encoder sọ awọn apẹrẹ igbi silẹ pẹlu awọn iyipada ti o tẹẹrẹ sinu ọkọọkan wiwaba, eyiti RVQ ṣe iyipada sinu awọn atọka iwe koodu tolera. Awoṣe ede Amunawa iwuwo fẹẹrẹ sọ asọtẹlẹ iṣeeṣe awọn ami ami wọnyi ati awọn koodu-iṣiro wọn, n bọlọwọ fun funmorawon siwaju fun ọfẹ. Oniwọntunwọnsi ikẹkọ ṣe atunṣe awọn ifunni gradient lati atunkọ, iwoye, ati awọn adanu ọta nitoribẹẹ ko si ọrọ kan ṣoṣo ti o jẹ gaba lori, eyiti o jẹ ki ikẹkọ idi-pupọ jẹ iduroṣinṣin ni iwọn iwọn bitrate ni kikun.
Ipa Ilana
Wiwọle ati arọwọto
O ṣe ilọsiwaju iraye si nipasẹ transcription, alaye, ati awọn atọkun ohun.
Iye owo ati isuna
Awọn ẹgbẹ Media le firanṣẹ ohun didan yiyara pẹlu awọn isuna-owo kekere.
Iyara ati iwọn
Awọn ọna ṣiṣe ti nkọju si alabara le ṣe ilana awọn ibaraẹnisọrọ sisọ ni iwọn nla.
Ojo iwaju ti EnCodec Audio funmorawon
EnCodec ti jẹ ami ami aifọwọyi fun ọpọlọpọ awọn awoṣe ohun afetigbọ ṣiṣi silẹ, ati pe awọn ọmọ rẹ n titari iṣotitọ ti o ga julọ ni awọn iwọn kekere, sitẹrio ni kikun ati atunkọ-orin, ati isọpọ tighter pẹlu ọrọ-si-ohun ati awọn olupilẹṣẹ ọrọ-si-orin. Reti isọdọmọ ti o gbooro ni ibaraẹnisọrọ bandiwidi kekere, ṣiṣanwọle akoko gidi, ati gẹgẹ bi ipele 'àmì ohun afetigbọ' ti o jẹ ki awọn faaji ara-ede nla ka ati kọ ohun.
Real-World imuse
Ohun afetigbọ fun Meta's MusicGen ati AudioGen awọn olupilẹṣẹ ọrọ-si-ohun
Compressing 24 kHz ọrọ to 1.5-6 kbps fun bandiwidi-lopin gbigbe
Fifi koodu sitẹrio 48 kHz pẹlu didara nitosi MP3 ni awọn iwọn biiti ti o ga pupọ
Ṣiṣẹ bi koodu ṣiṣi silẹ orisun orisun fun iwadii ati awọn opo gigun ti ML nipasẹ awọn aaye ayẹwo ti o tu silẹ
Awọn ewu & Awọn ọna iṣọ
ilokulo ohun ati awọn ewu afarawe ṣe pọ si nigbati igbanilaaye ba sonu.
Yiye le ju silẹ kọja awọn asẹnti, awọn ede-ede, tabi awọn agbegbe alariwo.
Ohun afetigbọ sintetiki le jẹ aṣiṣe fun ọrọ ododo laisi isamisi to yege.
Ilana Ilana imuse
Gba ifọkansi ti o fojuhan fun gbigba ohun, ti ẹda, ati ilotunlo.
Didara idanwo kọja awọn agbohunsoke oniruuru ati awọn ipo abẹlẹ.
Ṣetumo nigbati eniyan gbọdọ ṣe atunyẹwo tabi fọwọsi awọn abajade.
Aami ohun sintetiki ki o tọju awọn igbasilẹ provenance fun iṣiro.
Tesiwaju Ṣiṣawari
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the EnCodec Audio Compression quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Itọsọna atẹle
Awọn ifibọ ohun ati Ẹkọ Aṣoju
Awọn ibeere ti a beere nigbagbogbo
What is EnCodec Audio Compression?
EnCodec jẹ Meta's kodẹki ohun iṣan ti o ni ifaramọ ti o ga julọ ti o rọ ọrọ ati orin pọ ni awọn iwọn biiti kekere pupọ pẹlu idije didara ti awọn ọna kika ti o wuwo pupọ julọ. O ṣe pataki nitori pe o ṣe atilẹyin awọn eto ohun afetigbọ ode oni ati awọn ọkọ oju omi ni fọọmu orisun-ìmọ fun ẹnikẹni lati lo.
Ẹgbẹ wo ni o tu EnCodec silẹ?
EnCodec ti ṣe afihan nipasẹ Meta AI (FAIR) ni ọdun 2022 gẹgẹbi kodẹki ohun afetigbọ ohun aifọkanbalẹ ti o ga.
Ẹya afikun wo ni EnCodec ṣe afikun si ailagbara fun fun pọ diẹ sii lati awọn ami rẹ?
EnCodec ṣe ikẹkọ Amunawa kekere lati ṣe asọtẹlẹ awọn iṣeeṣe tokini ati awọn koodu-iṣiro wọn, ṣiṣe iyọrisi afikun funmorawon ailagbara lori oke RVQ.
Ni aijọju 6 kbps, didara EnCodec ni a royin bi afiwera si MP3 ni nipa bii bitrate wo?
EnCodec ni 6 kbps de didara ero-ara ti o jọra si MP3 ni 64 kbps, ere ṣiṣe nla kan.
Isoro wo ni EnCodec's 'balancer' yanju lakoko ikẹkọ?
Pẹlu ọpọlọpọ awọn adanu (atunṣe, spectral, adversarial), iwọntunwọnsi ṣe iwọn awọn gradients wọn nitorinaa ko si ipinnu kan ṣoṣo ti o jẹ gaba lori, ikẹkọ iduroṣinṣin.
Ọna titobi pataki wo ni EnCodec pin pẹlu SoundStream?
Bii SoundStream, EnCodec nlo iye iwọn fekito ti o ku lati sọ iyasọtọ awọn ifibọ koodu sinu awọn atọka iwe koodu tolera.