Jukebox Quiz

Answers and explanations appear as you take the quiz.

1. What makes Jukebox different from earlier symbolic music AI like systems that output MIDI?
   • It generates raw audio waveforms including singing voices, not just notes
   • It only generates drum patterns
   • It can only transcribe existing songs
   • It requires a human to play the melody first
2. Who released Jukebox and roughly when?
   • Google, in 2016
   • OpenAI, in 2020
   • Meta, in 2023
   • Spotify, in 2019
3. What was a major practical limitation of Jukebox?
   • It could only make 10-second clips
   • It refused to produce any vocals
   • Generation was extremely slow, taking hours to render a minute of audio
   • It needed a live orchestra as input
4. What core technique does Jukebox use to make long raw audio manageable for its Transformers?
   • It stores every song as a single pixel image
   • It only keeps the loudest note per second
   • It relies entirely on human-written sheet music
   • It converts audio into a short sequence of discrete codes using a VQ-VAE
5. What can you condition Jukebox's output on?
   • Genre, artist style, and lyrics
   • Only the tempo
   • Only the key signature
   • The listener's heart rate
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