I-Neural Audio Codecs
Amakhodekhi omsindo we-Neural asebenzisa ukufunda okujulile ukucindezela umsindo emifudlaneni emincane yamathokheni ahlukene futhi awakhe kabusha ngokwethembeka okuphezulu.
Uhlolojikelele
They both crush bandwidth for calls and streaming and provide the token vocabulary that audio language models speak.
I-Deep Dive
I-neural audio codec iyinethiwekhi ye-neural yesikhiphi khodi eqeqeshelwe ukucindezela umsindo nokuwakha kabusha. Isishumeki sekhodi sishintsha i-waveform ibe i-compact latent, i-quantizer ithwebula efihlekile kokufakwayo kuma-codebook afundiwe akhiqiza amathokheni ahlukene, futhi isiqophi siphinde sakhe i-waveform. Indlela eyinhloko i-Residual Vector Quantization (RVQ), esetshenziswa Google's SoundStream kanye ne-Meta's EnCodec: ama-codebook amaningana astakiwe, ngayinye ibhala iphutha elishiywe ngaphambilini, ukuze ukwazi ukuhwebelana nge-bitrate ngekhwalithi ngokusebenzisa i-codebook eyengeziwe noma embalwa. Lawa mamodeli afinyelela ikhwalithi emangazayo ngama-bitrate aphansi kakhulu, kwesinye isikhathi amakhilobhithi ambalwa ngomzuzwana, ehlula ama-codec akudala njenge-Opus noma i-MP3. Ngokudabukisayo, amathokheni ahlukene ayimamodeli afana ne-VALL-E ne-MusicGen akhiqiza.
I-Technical Insight
I-RVQ iyinhliziyo yomklamo. I-codebook yokuqala ithwebula ukulinganiselwa okumahhadla, futhi i-codebook ngayinye elandelayo ilinganisa iphutha eliyinsalela, ibeka imininingwane emincane. Ukuqeqeshwa kuhlanganisa ukulahlekelwa kokwakha kabusha, ngokuvamile kuzo zombili izizinda zesikhathi nezokubonwayo, okunobandlululo oluphikisayo olugcina okukhiphayo kuzwakala njengokwangempela, kanye nokulahlekelwa ukuzibophezela okugcina okukhiphayo kwesifaki khodi kuseduze nokufakiwe kwe-codebook ekhethiwe. Umphumela uba ukumelwa okusobala, okulandelanayo okucindezelwayo futhi okulula ukuthi isiguquli esingaphansi komfula siyimodeli.
I-Strategic Impact
Finyelela futhi ufinyelele
Ithuthukisa ukufinyeleleka ngokuloba, ukulandisa, nezixhumi ezibonakalayo zezwi.
Izindleko kanye nesabelomali
Amaqembu emidiya angathumela umsindo opholishiwe ngokushesha ngamabhajethi amancane.
Isivinini nesikali
Amasistimu abhekene nekhasimende angacubungula ukusebenzelana okukhulunyiwe ngesilinganiso esikhulu.
Ikusasa Le-Neural Audio Codecs
Ama-codec aguqulela kuma-bitrate aphansi ngisho nangaphansi ngama-codebook ambalwa, okwenza amathokheni omsindo ashibhe ukuze akhiqize amamodeli olimi. Ucwaningo luphokophele ekusakazeni, okuhlukile okubambezeleka okuphansi kokuxhumana kwesikhathi sangempela kanye namakhodekhi ahlanganisiwe aphatha inkulumo, umculo, nomsindo ojwayelekile kumodeli eyodwa. Njengoba umsindo okhiqizwayo uqhuma, i-codec iya ngokuya iphathwa njengethokheni eyabelwe kuyo yonke inkambu, ngakho ukuthuthukiswa lapha kuzwakala kuwo wonke amamodeli ombhalo-ube-inkulumo kanye nomculo owakhelwe phezulu.
Ukuqaliswa Komhlaba Wangempela
Izwi elicindezelayo lamakholi we-ultra-low-bandwidth nezinhlelo zokusebenza zesitayela se-walkie-talkie
Ihlinzeka ngefomethi yethokheni ehlukile ekhiqizwa yi-VALL-E, i-AudioLM, ne-MusicGen
Ukugcinwa okusebenzayo nokusakazwa komsindo wekhwalithi ephezulu ngengxenyana ye-MP3 bitrate
Ukudluliswa kwenkulumo kwesikhathi sangempela kuzimo zenethiwekhi ezinomsindo noma ezinzima
Izingozi & Guardrails
Ukusetshenziswa kabi kwezwi kanye nezingozi zokuzenza ongeyena ziyanda uma imvume ingekho.
Ukunemba kungase kwehle kuzo zonke izinhlobo zokuphimisela, izilimi zesigodi, noma izindawo ezinomsindo.
Umsindo wokwenziwa ungenziwa iphutha njengenkulumo eyiqiniso ngaphandle kokulebula okucacile.
Ukuqalisa Umhlahlandlela
Thola imvume esobala yokuthwebula izwi, ukuhlanganisa, nokusebenzisa kabusha.
Ikhwalithi yokuhlola kuzo zonke izipikha nezimo zangemuva.
Chaza ukuthi kunini lapho umuntu kufanele abuyekeze noma agunyaze okuphumayo.
Lebula umsindo wokwenziwa futhi ugcine amarekhodi atholakalayo ukuze aziphendulele.
Qhubeka Uhlole
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 Neural Audio Codecs 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
Umhlahlandlela olandelayo
I-SoundStream Neural Codec
Imibuzo evame ukubuzwa
What is Neural Audio Codecs?
Amakhodekhi omsindo we-Neural asebenzisa ukufunda okujulile ukucindezela umsindo emifudlaneni emincane yamathokheni ahlukene futhi awakhe kabusha ngokwethembeka okuphezulu. Zombili zichoboza umkhawulokudonsa wamakholi nokusakaza-bukhoma futhi zinikeza ulwazimagama lwethokheni olukhulunywa amamodeli olimi lomsindo.
Yiziphi izinto ezimbili ezenziwa i-neural audio codec ngokuyinhloko?
I-neural codec iyinethiwekhi ye-encoder-decoder ecindezela i-waveform ibe amathokheni ahlukene ahlangene bese iphinda yakhe umsindo osuka kuwo.
Iyiphi indlela yokulinganisa ephakathi kwama-codec afana ne-SoundStream ne-EnCodec?
I-RVQ inqwabelanisa ama-codebook amaningi lapho ngayinye ibhala khona iphutha lensalela yangaphambilini, ivumela i-tradeoff yekhwalithi-ngokuqhathaniswa ne-bitrate.
Ku-Residual Vector Quantization, i-codebook ngayinye elandelanayo ifaka ini?
I-codebook yokuqala inikeza ukulinganiselwa okumahhadla, futhi i-codebook ngayinye yakamuva ilinganisa iphutha elisele, yengeze imininingwane engconywana.
Kungani i-neural audio codec ibalulekile kumamodeli omsindo akhiqizayo?
Amathokheni ahlukene akhiqizwa ama-codec aba ulwazimagama amamodeli olimi olulalelwayo alubikezelayo futhi alukhiqize.
Ingabe ukusebenzisa ama-codebook amaningi ku-neural codec kuwathinta kanjani okukhiphayo?
Ukwengeza ama-codebook kuphakamisa i-bitrate kodwa kuthwebula imininingwane eyengeziwe, kuthuthukisa ukwethembeka; usebenzisa ikhwalithi yokuhweba embalwa ngokuminyanisa.