Mel-Frequency Cepstral Coefficients Quiz

Answers and explanations appear as you take the quiz.

1. What does the 'mel' in MFCC refer to?
   • The name of the inventor
   • A type of neural network layer
   • A unit of loudness in decibels
   • A perceptual frequency scale that mimics human pitch hearing
2. Which transform is applied last to produce the cepstral coefficients?
   • Discrete cosine transform
   • Fourier transform
   • Wavelet transform
   • Hilbert transform
3. Why are MFCCs relatively robust to a speaker's pitch?
   • They amplify the fundamental frequency
   • The DCT separates slow vocal-tract shape from rapid pitch harmonics
   • They ignore amplitude entirely
   • They convert audio to text first
4. Roughly how many MFCC coefficients are typically kept per frame?
   • Exactly 100
   • Around 1,000
   • Around 13
   • Just 1
5. Why is the log applied to the mel-band energies?
   • To convert to text
   • To remove all noise perfectly
   • To increase the sampling rate
   • To approximate how humans perceive loudness
6. In modern deep-learning audio systems, what often replaces handcrafted MFCCs?
   • Raw waveforms or log-mel spectrograms fed into learned front-ends
   • Plain text transcripts
   • Only the Fourier transform with no scaling
   • MIDI files
7. What do the MFCCs summarize after the discrete cosine transform?
   • The original waveform sample by sample
   • A compact representation of the log mel-spectrum’s broad shape
   • The exact phase of every frequency component
   • The semantic meaning of a spoken sentence
8. What information is lost when a signal is summarized by a small set of MFCCs?
   • The recording’s entire duration
   • Every estimate of average loudness
   • Fine spectral detail and phase information
   • The existence of a time axis