AudioLM Quiz

Answers and explanations appear as you take the quiz.

1. What core idea does AudioLM use to generate audio?
   • It treats audio as a sequence of tokens and predicts the next one, like a language model
   • It renders audio from sheet music
   • It samples random noise with no learning
   • It records and replays human performances
2. What is the role of 'semantic' tokens in AudioLM?
   • They capture fine timbre and recording noise
   • They capture long-term structure like phonetics and melody
   • They store the text transcript
   • They control the output file format
3. Which neural codec produces AudioLM's acoustic tokens?
   • MP3
   • WaveNet
   • SoundStream
   • Whisper
4. What does AudioLM require as training data?
   • Audio paired with detailed transcripts
   • Only MIDI files
   • Images of waveforms
   • Raw audio alone, with no transcripts
5. After AudioLM predicts semantic tokens, how does its hierarchical generation add acoustic detail?
   • It predicts coarse acoustic tokens, then fine acoustic tokens
   • It predicts fine acoustic tokens, then coarse acoustic tokens
   • It predicts coarse acoustic tokens before semantic tokens
   • It predicts acoustic tokens without a semantic-token stage
6. Which later system built directly on AudioLM's token-based approach?
   • Stable Diffusion
   • MusicLM
   • GPT-2
   • DALL-E
7. Which model does AudioLM use to produce its semantic token sequence?
   • A text-only GPT decoder trained on transcripts
   • The SoundStream waveform decoder
   • A self-supervised w2v-BERT representation model
   • A hand-authored MIDI score
8. What audio types did the AudioLM paper demonstrate continuing from an audio prompt?
   • Only environmental noise and no music
   • Spoken text transcripts and MIDI files only
   • Images and silent video clips
   • Speech and piano music