Music Tagging with Transformers Quiz

Answers and explanations appear as you take the quiz.

1. Why is music tagging treated as a multi-label problem?
   • Because a single track can carry several tags at once
   • Because each song has exactly one correct tag
   • Because tags are always numeric
   • Because only one genre exists per artist
2. In the Audio Spectrogram Transformer paper, what audio representation is split into patches?
   • A list of MIDI note names
   • A log-Mel spectrogram
   • A sequence of album-cover pixels
   • A transcription of the song’s lyrics
3. Why do music tagging models use a sigmoid output per tag instead of one softmax?
   • Sigmoids are faster to compute
   • Softmax cannot run on audio
   • Because tags can co-occur and aren't mutually exclusive
   • To reduce the number of tags
4. What is the role of self-supervised pretraining for models like MERT?
   • It labels songs by hand
   • It removes vocals from tracks
   • It only works on classical music
   • It learns strong audio representations from unlabeled clips before fine-tuning
5. Which dataset is commonly used to fine-tune and benchmark music taggers?
   • MagnaTagATune
   • MNIST
   • ImageNet
   • SQuAD
6. How do CLAP-style models enable zero-shot music tagging?
   • By memorizing every song
   • By aligning audio and text descriptions in a shared embedding space
   • By converting music to images
   • By requiring labels for every possible tag
7. Which metric can evaluate a music tagger’s ability to rank relevant tags highly?
   • Word error rate
   • Peak signal-to-noise ratio
   • Mean average precision
   • BLEU score
8. What does the musicnn repository provide?
   • A text-only sentiment classifier
   • A digital audio workstation
   • A collection of manually composed MIDI scores
   • Pretrained convolutional neural networks for music audio tagging