HejuruUbuyobozi bukurikira
Kwinjiza Amajwi no Kwiga Guhagararira
AI y'amajwi
Ubuyobozi bwa Audio AI
EnCodec ni __AIU_PROTECTED_13 __ 'kwizerwa cyane kwijwi ryamajwi ya codec yerekana imvugo numuziki kuri bitrate nkeya cyane hamwe nubwiza burwanya imiterere iremereye cyane.
Umva iki gitaboAI-byavuzwe · ~2 min
It matters because it underpins modern generative audio systems and ships in open-source form for anyone to use.
Yarekuwe na Meta AI mu 2022, EnCodec ikurikira igishushanyo mbonera cya SoundStream ya encoder, ibisigisigi bya vector bisigaye (RVQ), hamwe na decoder yatojwe kurangiza, ariko ikongeramo byinshi binonosoye. Ikoresha uburyo-busobekeranye bwa kodegisi ya kodegisi, ibipimo byinshi bya spekrogramu hamwe nigihe-cyo gutakaza ibyangiritse, hamwe nivangura rirwanya ubuziranenge. Umusanzu ugaragara ni moderi ntoya ya Transformer ishingiye kuri entropiya yerekana irushijeho gukanda kode igereranijwe nta gihombo, ikuramo bits ziyongereye nta gihombo cyiza. EnCodec irashiraho kandi iringaniza ihita ipima igihombo kinini cyamahugurwa ahiganwa kugirango bakomeze gushikama. Ikora 24 kHz monophonic na 48 kHz amajwi ya stereo, ikora kuri bitrate nka 1.5, 3, 6, na 12 kbps, naho kuri 6 kbps igera mubwiza ugereranije na MP3 kuri 64 kbps. Ibimenyetso byayo imbaraga __AIU_PROTECTED_13 __ 'MusicGen na AudioGen.
Tokenizing amajwi ya __AIU_PROTECTED_13 __ 'UmuzikiGen na AudioGen inyandiko-y-amajwi
Gucomeka imvugo ya 24 kHz kuri 1.5-6 kbps yo kwaguka kwagutse
Encoding ya 48 kHz umuziki wa stereo hamwe nubwiza hafi ya MP3 kuri bitrate yo hejuru cyane
Gukora nkisoko ifunguye-codec yubushakashatsi hamwe n'amajwi ya ML binyuze kuri bariyeri yasohotse
EnCodec isanzwe yerekana ibimenyetso byerekana amajwi menshi afungura amajwi, kandi abayikomokaho barimo gusunika ubudahemuka kuri bitrate yo hasi, stereo yuzuye hamwe nu rwego rwo kongera kwiyubaka, hamwe no guhuza imbaraga hamwe n’inyandiko-ku-majwi hamwe n’umwandiko-ku-muzika. Tegereza kwaguka kwagutse mu itumanaho rito, itumanaho-nyaryo, kandi nkurwego rusanzwe 'amajwi token' urwego rutuma ururimi runini-rwerekana imiterere yububiko rusoma kandi rwandika amajwi.
Ibaruwa y'amakuru y'ubuntu
Inkuru eshatu zizewe za AI buri gitondo cy'iminsi y'akazi, zanditse mu rurimi rworoshye. Ubuntu iteka ryose, nta kwamamaza.
Imeri imwe buri munsi y'akazi. Kwiyimura mu buryo bumwe. Ntitwigera tugurisha cyangwa ngo dusangire aderesi yawe.
Isuzume
Ibisubizo ako kanya kuri buri kibazo, n'icyemezo gishobora gusangirwa gifite ID yagenzurwa umaze gutsinda isomo.
Fasha uburezi bwa AI bw'ubuntu. AI Understanding ni umuryango utegamiye kuri leta wa 501(c)(3) — nta kwamamaza, nta kwishyura, iteka ryose. Tanga impano
EnCodec ni __AIU_PROTECTED_13 __ 'kwizerwa cyane kwijwi ryamajwi ya codec yerekana imvugo numuziki kuri bitrate nkeya cyane hamwe nubwiza burwanya imiterere iremereye cyane. Nibyingenzi kuko bishimangira sisitemu yamajwi yerekana amajwi hamwe nubwato muburyo bwugururiwe isoko kugirango umuntu wese akoreshe.
EnCodec yatangijwe na Meta AI (FAIR) mu 2022 nka codec ya majwi yizerwa cyane.
EnCodec itoza Transformer ntoya yo guhanura ibimenyetso byerekana ibimenyetso hamwe na arithmetic-code yabyo, igera kuri compression itagira igihombo hejuru ya RVQ.
EnCodec kuri 6 kbps igera kumiterere yibintu bisa na MP3 kuri 64 kbps, inyungu nini.
Hamwe nigihombo kinini (kwiyubaka, kwerekanwa, guhangana), kuringaniza ikuraho gradients zabo kuburyo ntampamvu nimwe yiganje, guhagarika amahugurwa.
Kimwe na SoundStream, EnCodec ikoresha umubare wibisigisigi bya vector kugirango isobanure neza kodegisi yerekana indangagaciro za codebook.
Komeza wige
Abandi bayobozi batoranijwe kuriyi ngingo
HejuruUbuyobozi bukurikira
Kwinjiza Amajwi no Kwiga Guhagararira
AI y'amajwi