Ngụkọta akụkụ nke vector fọdụrụnụ
Residual vector quantization (RVQ) bụ usoro na-atụgharị ntinye ọdịyo na-aga n'ihu ka ọ bụrụ nchịkọta kọmpụta nke koodu pụrụ iche site n'ịgụta njehie fọdụrụ ugboro ugboro.
Nchịkọta
It matters because it is the engine behind modern neural codecs like SoundStream and EnCodec and the tokenizer for generative audio.
Ime miri emi
Plain vector quantization (VQ) na-edochi vector na-aga n'ihu na ntinye kacha nso na akwụkwọ koodu mmụta, mana otu koodu koodu nke zuru oke maka ịdị mma dị elu ga-achọ ntinye ọnụ ọgụgụ dị ukwuu nke mbara igwe. RVQ na-edozi nke a site n'itinye ọtụtụ akwụkwọ koodu obere obere. Nke mbụ codebook na-emepụta a keadighi rarii approximation; ị na-ewepụ ya ka ị nweta njehie fọdụrụnụ, jiri akwụkwọ koodu nke abụọ were gụpụta nke fọdụrụnụ, wepụ ọzọ, wee gaa n'ihu maka usoro N. Koodu ikpeazụ bụ ndepụta indices ahọpụtara n'ofe niile, na nrụgharị ahụ bụ nchikota vectors codebook niile ahọpụtara. Nke a na-emepụta nnukwu koodu codebook dị irè n'ime ọtụtụ ndị dị nta, na-egbutu ebe nchekwa n'ụzọ dị egwu ma na-agbakọ ma na-ahapụ ọnụ ọgụgụ bitrate naanị site na iji usoro ma ọ bụ ole na ole. Mwepu nke Quantizer n'oge ọzụzụ na-eme ka akwụkwọ codebook mbụ buru ozi kachasị, na-eme ka mmebi dị mma mara mma.
Nghọta nka nka
Nkeji ọ bụla na-enyocha akwụkwọ agbataobi kacha nso n'akwụkwọ koodu ya na ihe fọdụrụ ugbu a, a na-amụtakwa akwụkwọ koodu site na nwelite nkezi na-agagharị agagharị gbakwunyere mfu nkwa ka mpụta koodu wee nọrọ nso na ndenye ahọpụtara. Site na ntinye M nke K nke ọ bụla, RVQ na-anọchite anya K-to-the-M ngwakọta dị irè site na iji naanị M times K echekwara vectors na M times log2(K) ibe n'ibe kwa etiti, dị ọnụ ala karịa otu nnukwu codebook.
Mmetụta atụmatụ
Nweta na iru
Ọ na-eme ka nnweta ya dịkwuo mma site na ndegharị, ịkọ akụkọ, na ntụgharị olu.
Ọnụ ego na mmefu ego
Ndị otu mgbasa ozi nwere ike ibubata ọdịyo a na-egbu maramara ngwa ngwa site na iji obere mmefu ego.
Ọsọ na ọnụ ọgụgụ
Sistemụ na-eche ihu ndị ahịa nwere ike hazie mkparịta ụka n'ọtụtụ buru ibu.
Ọdịnihu nke Mbelata Vector Quantization
RVQ aghọwo ọkọlọtọ discretization oyi akwa na-ejikọta ihe nnochi anya akwara ozi na-aga n'ihu na ụdị mmepụta ihe na-adabere na token, na nhazigharị na-aga n'ihu: ntinye akwụkwọ koodu ka mma iji zere ndenye 'nwụrụ anwụ', akwụkwọ codebook nke nwere ike na nke dị ala, na usoro nhazi akara ngosi pụtara ìhè. Na-abụghị ọdịyo, otu ihe mpụta-stacking echiche na-agbasa na oyiyi na video tokenizers, n'ọnọdu RVQ dị ka a n'ozuzu àkwà mmiri n'etiti na-aga n'ihu encoders na asụsụ-ụdị ụdị usoro generators.
Mmejuputa n'ezie n'ụwa
Ihe ntinye koodu na-akpachapụ anya n'ime SoundStream, EnCodec, na DAC neural codecs
Na-emepụta token ọdịyo nwere ọkwa nke AudioLM na MusicLM na-ewepụta
Ịkwalite codec's bitrate elu ma ọ bụ ala site n'ịgbalite ọkwa quantizer karịa ma ọ bụ dị nta
Na-echikota ihe ntinye akụkụ dị elu na nweghachi na sistemu nchekwa site na iji akwụkwọ koodu ekpokọtara
Ihe ize ndụ & okporo ụzọ nche
Iji olu eme ihe na ihe egwu mpụta ga-abawanye mgbe nkwenye na-efu.
Izi ezi nwere ike ịdaba n'ofe ụda olu, olumba ma ọ bụ gburugburu mkpọtụ.
Enwere ike imehie ọdịyo sịntetik dị ka ezigbo okwu na-enweghị akara doro anya.
Map mmejuputa
Nweta nkwenye doro anya maka ijide olu, imechi, na ijigharị.
Nwale ogo n'ofe ndị na-ekwu okwu dị iche iche yana ọnọdụ ndabere.
Kọwaa mgbe mmadụ ga-enyocha ma ọ bụ kwado nsonye.
Deba aha ọdịyo sịntetik ma debe ndekọ ihe ndekọ maka ịza ajụjụ.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Mgbakwụnye Ọkà Okwu X-Vector
Ajụjụ a na-ajụkarị
What is Residual Vector Quantization?
Residual vector quantization (RVQ) bụ usoro na-atụgharị ntinye ọdịyo na-aga n'ihu ka ọ bụrụ nchịkọta kọmpụta nke koodu pụrụ iche site n'ịgụta njehie fọdụrụ ugboro ugboro. Ọ dị mkpa n'ihi na ọ bụ injin dị n'azụ codecs neural ọgbara ọhụrụ dị ka SoundStream na EnCodec na tokenizer maka ọdịyo mmepụta.
Kedu ihe koodu koodu ọ bụla na-esochi na RVQ na-atụle?
Mgbe akwụkwọ koodu nke mbụ nyechara atụmatụ siri ike, RVQ na-ewepụ ya wee were akwụkwọ koodu na-esote gụgharịa ihe fọdụrụnụ, na-emegharị n'ofe ọkwa.
Kedu ihe kpatara eji RVQ kama ịbụ otu nnukwu koodu koodu?
Otu akwụkwọ koodu dị mma zuru oke maka ịdị mma dị elu ga-aba nnukwu uru; ịtụkọ akwụkwọ koodu M nke ndenye K na-enye mkpokọta K^M nwere obere nchekwa.
Kedu ka esi emepụta nwughari ikpeazụ site na koodu RVQ?
Nrụgharị ahụ bụ nchikota nke vectors codebook ahọpụtara n'ofe niile, nke ọ bụla na-edozi ihe fọdụrụ nke ndị mbụ.
Site na ntinye M nke K nke ọ bụla, nchikota koodu ole dị irè ka RVQ nọchiri anya ya?
Ịhọrọ otu n'ime ndenye K na ọkwa nke onwe nke ọ bụla na-enye K^M nchikota enwere ike, ebe ị na-echekwa naanị M*K vectors.
Kedu ihe bụ ebumnuche nke 'mfu nraranye' mgbe ị na-azụ akwụkwọ koodu RVQ?
Mfu nkwa ahụ na-ata onye ngbanwe ahụ ahụhụ mgbe ihe nrụpụta ya si na vectors codebook ahọpụtara, na-eme ka ọnụọgụgụ kwụsie ike; codebooks n'onwe ha na-emelitekarị site na nkezi na-agagharị agagharị.