AudioLM
AudioLM Google kaadaru gëstu la buy defar ay audio yu dëggu — wax wala music piano — ci jëfandikoo son bi ni làkk, ba noppi diko wax luy waaja am.
Résumé
It matters because it showed you can produce coherent, natural-sounding audio continuations without any text transcript or musical score.
Plongeur bu xóot
Google moo ko dugal ci 2022, AudioLM dafay soppi defar audio ni jafe-jafe modeling làkk: dafay soppi forme onde yu ñor yi ci ay token yu diskret ba noppi di wax luy token bi ci topp, ni ko modelu bind di waxee baat bi ci topp. Kaf gi gëna am solo mooy hierarchie xeetu token yi. Jeton 'semantik' (yu bawoo ci xeetu w2v-BERT) dafay jàpp jëmmal bu yàgg - fonetik, sintaks, melodi - fekk jeton 'akustik' (yu bawoo ci kodek neural SoundStream) dafay jàpp ay detay yu rafet yu melni dàntite waxkat, timbre, ak anam enregistrement. Su AudioLM njëkkee wax luy waaja am ci token semantik yi, ba noppi nga def ci token akustik yi, AudioLM dafay defar ay wéy yuy wéy di lëkkaloo ci diiru segond yu bari, boole ci baña yàq baat bi wala jumtukaay bi njëkk. Soo ko joxee ay segond yu néew mu wax, mu wéy di wax ak baat bimu waxoon; buñu ko jox piano, dafay improvise ci benn stil.
Gis-gis xarala
AudioLM dañu ko tàggat ci audio kese — amul ay transkripsioŋ. SoundStream dafay komprime audio bi ci ay jeton akustik jaaraleko ci kantite vecteur residuel, fekk w2v-BERT dafay joxe ay jeton semantik yu dëgërul. Benn xeetu làkku Transformer dafay wax luy waaja am ci ay etape: semantik bi njëkk ngir structure, ginaaw ga ay token akustik yu dëgër ak yu rafet ngir tabaxaat bu baax. Dekodeeru SoundStream dafay mujjee soppi token yiñ séentuwoon ci forme vague, mu am audio buy tëye baat bi ak prosodie bi.
njeextalu pexe
Dugg ak yegg
Dafay gëna yombal jëfandikoo gi jaaraleko ci transkripsioŋ, nettali ak interfaasu baat.
Njëgg ak budget
Ekipu mejaa yi mën nañu yónnee audio bu leer ci anam wu gëna gaaw te seen xaalis gëna néew.
Gaawaay ak yaatuwaay
Sistem yiy jàkkarloo ak kiliyaan bi mën nañu def waxtaan ci anam wu gëna yaatu.
Ëlëgu AudioLM
Reset bu AudioLM bi lalu ci token moo nekkoon fundamaasu sistem yu ci topp: __AIU_PROTECTED_11_ xalaati AudioLM dugal nañu ko ci MusicLM ngir bind-ci-musik ak SoundStorm ngir gëna gaaw ci génne, ci noonu la barab bu gëna yaatu bi leegi di boole token semantik ak akustik, efekti son, music, ci biir. Xaarandil lu gëna gaaw, defar ci jamono dëgg, génne gu gëna gudd te méngoo, ak doxal multimodal fu bind wala yeneen siñaal di doxal model yuñ tàggat ci audio. Pexe yooyu dañuy gëna yokk jafe-jafe yi ci wàllu klonaasu baat ak deepfake audio.
Doxal ci àdduna dëgg
Wéyal ab clip diskuur bu gàtt ci benn baat ak intonaasioŋ bu waxkat bi te du am benn transkripsioŋ
Defar music piano bu bees bu méngoo ak stil ab enregistrement bu gàtt
Dafay nekk yaxu ndigg ci defar audio ngir sistem bind-ci-musik yu melni MusicLM
Gëstu ci synthese kàddu buy denc prosody ak enregistrement akustik ci misaal
Risk yi ak balustrade yi
Jëfandikoo baat ci anam wu jaarul yoon ak niru ak nit dafay gëna yokk sudee nanguwul.
Jaar-jaar mën na wàññeeku ci aksan yi, dialect yi wala barab yu bari xumbaay.
Audio synthetik mën nañu ko jaawale ak wax ju dëggu sudee amul etiket bu leer.
Roadmap ngir samp gi
Wutal ndigal bu leer ngir jàpp baat bi, klone ko ak jëfandikoowaat ko.
Saytu kalite ci kàddukat yu bari ak anam yu bari ci ginaaw.
Mandargal kañ la nit wara xoolaat wala nangu ay génne.
Etiketu audio synthetik te nga denc dokimaa ci fimu bawoo ngir mëna lim.
Weyal di banneexu
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 AudioLM 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
Gis bi ci topp
Baatu-caabi Spotting ak Wake Words
Laaj yi ñuy faral di laaj
What is AudioLM?
AudioLM Google kaadaru gëstu la buy defar ay audio yu dëggu — wax wala music piano — ci jëfandikoo son bi ni làkk, ba noppi diko wax luy waaja am. Dafa am solo ndax dafa wane ni mën nga defar ay wéyal audio yu méngoo, di sone bu baax te doo jëfandikoo benn mbind wala partition musical.
Ban xalaat bu am solo la AudioLM di jëfandikoo ngir defar audio?
AudioLM dafay soppi jëmmu vague yi ci ay token yu wuute ba noppi di leen wax luy waaja xew ci anam wu ñuy toppalante, di jëfandikoo jegewaale token bi ci topp ci xeetu làkk yi ci son bu ñor.
Lan mooy wareefu jeton 'semantik' ci AudioLM?
Jetons semantik (yu bawoo ci w2v-BERT) dañuy kode structure ak cohérence bu kawe, waaye jetons akustik yi dañuy jëfandikoo ay detay audio yu rafet.
Ban codec neuronal mooy defar jeton akustik yu AudioLM?
SoundStream dafay jëfandikoo kantite vecteur residuel ngir kompresse audio ci ay jeton akustik ginaaw ga mu dekode jeton yiñ séentu ci forme onde.
Lan la AudioLM soxla ci done tàggat?
Benn ci màndarga yi gëna fës mooy AudioLM dafay tàggat ci audio kese te amul benn etiketu bind, di jàng structure ci son bi.
Lan moo waral AudioLM di wax luy waaja am ci jeton semantik yi balaa jeton akustik yi?
Defar structure bu dëgërul njëkk mooy tax génne gi gëna dëppoo ci diir bi; jetons akustik yi deñu leen di def ci structure bi ngir yokk ay detay yu fidélité bu kawe.