learn.ifs.name

Phonetics and Speech Feature Extraction

Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 15

18 cards in “15.6 MFCC” · whole deck

15.6 MFCC

  1. What does MFCC stand for?
    Mel frequency cepstral coefficients.
  2. What are the 39 dimensions of an MFCC vector?
    12 cepstral coefficients + 1 energy; 12 delta cepstral + 1 delta energy; 12 double-delta cepstral + 1 double-delta energy.
  3. Why is the cepstrum useful for phone detection?
    It separates source and filter. Glottal source characteristics (F0, pulse details) don't distinguish phones; the vocal tract filter (articulator positions) does.
  4. What does it mean to deconvolve source and filter?
    Separate them so features show only the vocal tract filter.
  5. What is the cepstrum, informally?
    The spectrum of the log of the spectrum.
  6. When the log spectrum is treated as a “pseudo-signal”, what do its high- and low-frequency components represent?
    High-frequency ripple: the harmonics of F0 (the source). Low-frequency components: the spectral envelope / formant structure (the filter).
  7. Where does the word cepstrum come from?
    Reversing the first four letters of spectrum.
  8. What is the unit of the cepstrum's x-axis, and why?
    Samples: taking the spectrum of the log spectrum leaves the frequency domain and returns to the time domain.
  9. In the example cepstrum, what does the large peak around 120 represent? The lower values?
    The peak corresponds to F0 (glottal pulse). Lower values represent the vocal tract filter.
  10. Which cepstral values are used for phone detection vs pitch detection?
    Lower cepstral values for phones; higher cepstral values for pitch.
  11. How many cepstral values does MFCC extraction keep, and why?
    The first 12; they represent only the vocal tract filter, cleanly separated from the glottal source.
  12. What useful statistical property do cepstral coefficients have?
    Their variances tend to be uncorrelated, unlike spectral coefficients at different frequency bands.
  13. Formal definition of the cepstrum?
    The inverse DFT of the log magnitude of the DFT of a signal.
  14. What is the 13th MFCC feature and how is it computed?
    Frame energy: \(\text{Energy} = \sum_{t=t_1}^{t_2} x^2[t]\), the sum over time of the samples' power.
  15. Why is energy a useful feature?
    It cues phone identity, e.g. vowels and sibilants have more energy than stops.
  16. What are delta and double delta features?
    Delta (velocity): change between frames in each of the 13 features. Double delta (acceleration): change between frames in the deltas.
  17. Why add delta features?
    Changes over time — like formant slopes at transitions, or stop closure → burst — are useful cues for phone identity.
  18. How are delta features computed in practice?
    Simply by frame differences, but commonly by fitting a polynomial and taking its first and second derivatives.

Source