Filterbank and PLP Features Quiz

Answers and explanations appear as you take the quiz.

1. Why do speech filterbanks use mel- or Bark-spaced bands instead of uniformly spaced hertz?
   • The scales allocate finer frequency spacing at low frequencies and coarser spacing at high frequencies, approximating aspects of hearing
   • They make files smaller only
   • They remove all background noise
   • They guarantee perfect speech recognition
2. What does the linear prediction step in PLP accomplish?
   • It increases the volume
   • It fits a smooth spectral envelope capturing vocal-tract shape
   • It adds reverb
   • It converts audio to text directly
3. Which feature type is the dominant input for modern deep speech recognition models?
   • Raw MIDI files
   • Handwritten transcripts
   • Log-mel filterbank features
   • JPEG spectrograms
4. How do MFCCs differ from raw log-mel filterbank features?
   • MFCCs use no filters at all
   • MFCCs are measured in decibels only
   • MFCCs require a microphone array
   • MFCCs add a cosine transform to decorrelate the filterbank outputs
5. What perceptual element does PLP include that basic mel filterbanks do not?
   • An equal-loudness curve and cube-root intensity compression
   • A built-in language model
   • Automatic translation
   • Stereo panning
6. Why might a deep model use log-mel features or raw waveforms instead of PLP?
   • PLP is illegal to use
   • The model can learn useful task-specific representations without every hand-designed PLP step
   • Raw audio is always smaller
   • PLP only works for music
7. What is a typical log-mel filterbank feature for one speech frame?
   • A word-level translation
   • A single pitch label for the whole recording
   • The logarithm of spectral energy pooled through a mel-spaced filter
   • A pixel color from a spectrogram image
8. Which order best describes the main PLP processing steps?
   • Translation, tokenization, and decoding
   • Stereo panning, reverb, and compression
   • Image segmentation followed by color correction
   • Critical-band analysis, equal-loudness weighting, loudness compression, then linear prediction