CREPE Pitch Estimation Quiz

Answers and explanations appear as you take the quiz.

1. What does CREPE estimate from an audio signal?
   • The fundamental frequency (pitch)
   • The loudness envelope
   • The tempo in BPM
   • The instrument's brand
2. CREPE is designed primarily for what kind of audio?
   • Full orchestral mixes
   • Monophonic (single-note) audio
   • Stereo field recordings
   • Compressed podcasts only
3. How does CREPE frame the pitch-estimation task internally?
   • As regression of a single number
   • As clustering of waveforms
   • As classification over many pitch bins
   • As reinforcement learning
4. What does CREPE take as its direct input?
   • A spectrogram image
   • A MIDI file
   • Hand-crafted autocorrelation features
   • Raw time-domain audio frames
5. Approximately how far apart are CREPE's pitch bins?
   • 20 cents
   • 1 semitone
   • 1 octave
   • 100 Hz
6. What did the CREPE paper report about performance against pYIN?
   • CREPE was proven more accurate on every audio signal
   • CREPE performed equally or better in the evaluated settings, including tests of noise robustness
   • CREPE requires no training data
   • CREPE only works on clean speech
7. How many waveform samples does an original CREPE frame contain?
   • 360 samples
   • 128 samples
   • 1,024 samples
   • 16,000 samples
8. How does CREPE refine a frequency estimate around the strongest pitch bin?
   • By averaging every bin equally across the full range
   • By selecting the lowest activation
   • By mapping only to the nearest semitone label
   • By taking a local weighted average of nearby activations