Guide-linked quiz · Medium Level

Weakly Supervised Sound Event Detection Quiz

Test weak clip labels versus strong event boundaries, pooling, overlapping sounds, timing evaluation, and synthetic-to-real shift.

Question 1 of 8

A clip is tagged “doorbell” with no timestamp. Why is that weak supervision for detection?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. A clip is tagged “doorbell” with no timestamp. Why is that weak supervision for detection?
    • The clip contains no waveform
    • The label gives presence but not event onset or offset
    • A doorbell cannot be recognized acoustically
    • Only one event can occur in ten seconds
  2. What additional data are needed to score a model’s proposed event timing directly?
    • Only the clip’s class tag
    • Only a model confidence score
    • Only the recording file size
    • Human-marked or otherwise trusted onset and offset references
  3. A model predicts “doorbell” correctly but at the wrong second. What does that show?
    • Clip classification and temporal localization are distinct
    • The clip label must be wrong
    • The event source was certainly isolated
    • The recording cannot contain overlapping events
  4. Why can a clip contain several positive event labels?
    • Each clip must be reduced to one class
    • The task assumes non-overlapping events
    • Sounds can overlap or occur at different moments in one recording
    • Only a single onset can be scored per recording
  5. How can a frame-level model train from a clip-level tag?
    • Assume every frame has the event label
    • Aggregate frame scores into a clip score for comparison with the tag
    • Ignore the sound waveform
    • Use the test-set event times as training labels
  6. What risk arises from training on precisely labeled synthetic mixtures?
    • The generated acoustics may not match real recordings
    • Synthetic examples have no timestamps
    • Real audio cannot contain multiple sources
    • A known mixture guarantees deployment accuracy
  7. Why specify timing tolerance or overlap rules in evaluation?
    • Every detector has a universal natural cutoff
    • A weak clip label defines the exact onset
    • Different matching rules leave event scores unchanged
    • Boundary scores depend on what counts as a match
  8. Which practice best supports a deployed detector after training with weak labels?
    • Report only clip-tagging accuracy
    • Assume every untagged sound is absent
    • Test false alarms, missed sounds and event timing on representative audio
    • Keep only the most attractive demo

Keep testing yourself

More quizzes picked for your level.

View all quizzes