AudioGen Text-to-Audio Synthesis Quiz

Answers and explanations appear as you take the quiz.

1. What task does the cited AudioGen study address?
   • Only exact transcript-to-speech rendering
   • Only image classification
   • Generating general audio conditioned on descriptive text
   • Only converting every audio event into MIDI notation
2. What is the first stage in AudioGen's pipeline?
   • An audio autoencoder that turns waveforms into discrete tokens
   • A text-to-image model
   • A speech recognizer
   • A video encoder
3. What type of model predicts the audio tokens?
   • A convolutional image classifier
   • A decision tree
   • A simple linear regressor
   • An autoregressive Transformer language model
4. What does AudioGen’s original mixed-source augmentation do?
   • Only concatenates unmodified waveform clips end to end
   • Mixes audio samples and concatenates their corresponding text captions
   • Deletes both source captions
   • Changes every sound into a silent recording
5. Which larger Meta library includes AudioGen?
   • AudioCraft
   • PyTorch Lightning
   • TensorFlow Hub
   • Hugging Face Diffusers
6. How is a final waveform produced from AudioGen's generated tokens?
   • The tokens are printed as text
   • They are emailed to a server
   • They are rendered as an image
   • The codec decoder converts tokens back into audio
7. Which two model contributions are combined by classifier-free guidance in the original AudioGen formulation?
   • Only the left and right display colors
   • Text-conditional and unconditional log-probability contributions
   • Only two unrelated image filenames
   • A required human transcript and a password
8. What additional input defines AudioGen’s studied audio-continuation setup?
   • Only a blank file extension
   • A mandatory complete future waveform as the answer
   • An initial acoustic audio prefix
   • Only a screenshot of the source code