Ibisigisigi bya Vector
Ibisigisigi bya vector bisigaye (RVQ) nubuhanga buhindura amajwi adahwema gushira mubice byegeranye bya code ya discret mukubara inshuro nyinshi amakosa asigaye.
Incamake
It matters because it is the engine behind modern neural codecs like SoundStream and EnCodec and the tokenizer for generative audio.
Kwibira cyane
Umubare wuzuye wa vector (VQ) usimbuza icyerekezo gikomeza hamwe nicyinjira hafi yigitabo cyize, ariko codebook imwe ihagije bihagije kubwiza buhanitse yakenera umubare munini winyenyeri. RVQ ikemura ibi mugukurikirana kodegisi ntoya. Igitabo cya mbere codebook itanga igereranya rito; ukuramo kugirango ubone ikosa risigaye, gereranya ibyo bisigara hamwe na codebook ya kabiri, ongera ukuremo, hanyuma ukomeze kuri N ibyiciro. Kode ya nyuma ni urutonde rwibintu byatoranijwe mu byiciro byose, kandi kwiyubaka ni igiteranyo cya codebook zose zatoranijwe. Ibi byerekana kodegisi nini cyane mubuto buto, igabanya cyane kwibuka no kubara mugihe ureka igipimo cya bitrate ukoresheje ibyiciro byinshi cyangwa bike. Quantizer yataye mugihe cyamahugurwa ituma kode ya kare itwara amakuru menshi, igafasha kwangirika kwiza.
Ubushishozi
Buri cyiciro gikora hafi-umuturanyi ushakisha hejuru ya codebook kuri ibisigisigi byubu, kandi codebooks yigishwa hamwe na exponential-igenda-igereranya igezweho hiyongereyeho igihombo cyo kwiyemeza bityo ibisubizo bya kodegisi biguma hafi yibyatoranijwe. Hamwe na M ibyiciro bya K byinjira buriwese, RVQ yerekana K-kuri-M ikora neza ikoresheje inshuro M gusa K yabitswe na M inshuro log2 (K) bits kumurongo, bihendutse cyane kuruta codebook imwe nini.
Ingaruka z'Ingamba
Kugera no kugera
Itezimbere kugerwaho binyuze mu kwandukura, kuvuga, no guhuza amajwi.
Igiciro na bije
Amatsinda yibitangazamakuru arashobora kohereza amajwi yihuse hamwe na bije nto.
Umuvuduko n'igipimo
Sisitemu ireba abakiriya irashobora gutunganya imikoranire ivugwa murwego runini.
Ejo hazaza h'ibisigisigi bisigaye
RVQ yahindutse urwego rusanzwe rwa discretisation ihuza imitekerereze ikomeza yerekana imiterere yibyara ibyara, kandi binonosorwa birakomeza: gukoresha neza codebook kugirango wirinde ibyanditswe 'byapfuye', kodegisi yibikoresho kandi biri munsi, hamwe nibisobanuro byerekana ibimenyetso. Hanze y'amajwi, igitekerezo kimwe gisigaye-gikwirakwiza gikwirakwira kumashusho na videwo yerekana amashusho, ugashyira RVQ nkikiraro rusange hagati ya kodegisi ikomeza hamwe nururimi-rwerekana-imiterere-yimikorere ikurikirana.
Gushyira mu bikorwa Isi
Gutandukanya kodegisi yashizwemo imbere muri SoundStream, EnCodec, na DAC codecs
Gukora amajwi yerekana amajwi AudioLM na MuzikaLM itanga hejuru
Gupima codec ya bitrate hejuru cyangwa hepfo mugukora ibyiciro byinshi cyangwa bike
Gucomeka murwego rwohejuru rwinjizwamo muri sisitemu yo kugarura no kubika ukoresheje codebook zegeranye
Ingaruka & Kurinda
Gukoresha nabi amajwi no kwigira ibyago byiyongera mugihe uruhushya rubuze.
Ukuri kurashobora kugabanuka hejuru yimvugo, imvugo, cyangwa urusaku rwibidukikije.
Amajwi yubukorikori arashobora kwibeshya kumvugo yukuri nta kirango gisobanutse.
Igishushanyo mbonera
Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.
Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.
Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.
Andika amajwi yubukorikori kandi ugumane inyandiko zerekana kubazwa.
Komeza Ubushakashatsi
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 Residual Vector Quantization 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
Ubuyobozi bukurikira
X-Vector Speaker Embeddings
Ibibazo bikunze kubazwa
What is Residual Vector Quantization?
Ibisigisigi bya vector bisigaye (RVQ) nubuhanga buhindura amajwi adahwema gushira mubice byegeranye bya code ya discret mukubara inshuro nyinshi amakosa asigaye. Nibyingenzi kuko ni moteri inyuma ya codecs igezweho ya kijyambere nka SoundStream na EnCodec hamwe na tokenizer kumajwi atanga.
Niki buri gitabo cyakurikiranye muri RVQ kigereranya iki?
Nyuma ya codebook yambere itanga igereranya rito, RVQ irayikuramo ikagereranya ibisigisigi bisigaye hamwe na codebook ikurikira, igasubiramo ibyiciro.
Kuki ukoresha RVQ aho gukoresha code imwe nini?
Kode imwe imwe ihagije kubwiza bwo hejuru yaba nini idasanzwe; gutondekanya M codebooks ya K ibyinjira biha K ^ M guhuza hamwe nububiko buke cyane.
Nigute kwiyubaka kwa nyuma bikorwa kuva kode ya RVQ?
Kwiyubaka ni igiteranyo cya codebook yatoranijwe mu byiciro byose, buri kimwe gikosora ibisigisigi byabanjirije.
Hamwe na M ibyiciro bya K byinjira buriwese, kangahe kode ihuza RVQ ihagarariye?
Guhitamo imwe muri K ibyanditswe kuri buri cyiciro cyigenga gitanga K ^ M ibishoboka byose, mugihe ubitse M * K gusa.
Niyihe ntego isanzwe y 'igihombo cyo kwiyemeza' mugihe uhugura kode ya RVQ?
Igihombo cyo kwiyemeza gihana kodegisi mugihe ibisubizo byayo biva kumurongo watoranijwe wa codebook, bigatuma umubare uhagaze neza; codebooks ubwayo ikunze kuvugurura ikoresheje impuzandengo yimuka.