Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Jukebox adanwo

Review Jukebox’s discrete-code music-generation method, its metadata conditioning and training data, and why waveform generation was computationally slow.

Jẹmọ itọsọna awọn ọnaJukebox
Ibeere 1 ti 8

Kini o jẹ ki Jukebox yatọ si orin aami iṣaaju AI bii awọn eto ti o ṣe agbejade MIDI?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini o jẹ ki Jukebox yatọ si orin aami iṣaaju AI bii awọn eto ti o ṣe agbejade MIDI?
    • O ṣe agbekalẹ awọn igbi ohun afetigbọ aise pẹlu awọn ohun orin, kii ṣe awọn akọsilẹ nikan
    • O ṣe ipilẹṣẹ awọn ilana ilu nikan
    • O le ṣe igbasilẹ awọn orin ti o wa tẹlẹ nikan
    • O nilo eniyan lati kọ orin aladun ni akọkọ
  2. Tani o tu Jukebox silẹ ati ni aijọju nigbawo?
    • Google, ni ọdun 2016
    • OpenAI, ni ọdun 2020
    • Meta, ni ọdun 2023
    • Spotify, ni ọdun 2019
  3. Kini aropin ilowo pataki ti Jukebox?
    • O le ṣe awọn agekuru iṣẹju-aaya 10 nikan
    • O kọ lati gbe awọn ohun orin jade
    • Iran ti lọra pupọ, o gba awọn wakati lati mu ohun iṣẹju kan ṣe
    • O nilo ẹgbẹ orin laaye bi titẹ sii
  4. Ilana mojuto wo ni Jukebox lo lati jẹ ki ohun afetigbọ aise gun le ṣakoso fun Awọn Ayirapada rẹ?
    • O tọju gbogbo orin bi aworan ẹbun kan
    • O ntọju akọsilẹ ti o pariwo nikan fun iṣẹju-aaya
    • O gbarale patapata lori orin dì ti eniyan kọ
    • O ṣe iyipada ohun ohun sinu ọna kukuru ti awọn koodu ọtọtọ nipa lilo VQ-VAE kan
  5. Kini o le ni ipo iṣelọpọ Jukebox lori?
    • Oriṣi, ara olorin, ati awọn orin
    • Iwọn akoko nikan
    • Ibuwọlu bọtini nikan
    • Iwọn ọkan ti olutẹtisi
  6. What duration of coherent music did the Jukebox paper report the combined model could generate?
    • Only a few milliseconds
    • Up to multiple minutes
    • Exactly one hour in every sample
    • Only single isolated notes
  7. What role do lyrics play in Jukebox’s conditioning setup?
    • Lyrics are used only as evaluation labels after generation
    • Lyrics are converted to MIDI and ignored by the audio model
    • Unaligned lyrics can make generated singing more controllable
    • Lyrics are not part of the paper’s conditioning
  8. Approximately how many songs were used in Jukebox’s training collection?
    • About 1,200 songs
    • About 12,000 songs
    • About 120 million songs
    • About 1.2 million songs