Spectral

Reporting item 5 — spectral centroid and item 7 — spectral slope β.

from debussy.spectral import spectral_centroid_hz, spectral_slope

spectral_centroid_hz(y, fs, librosa) -> float

Mean of librosa.feature.spectral_centroid across frames — the spectral centre of gravity, in Hz. Rises with high-frequency energy, so it tracks perceived brightness and correlates with (but is not) sharpness.

spectral_slope(y, fs) -> dict

Least-squares slope of log₁₀ power against log₁₀ frequency, fitted over 50 Hz to fs/2 on a Hann-windowed segment of up to 65 536 samples.

Returns {"beta": float | None, "band_lo_hz": 50.0, "band_hi_hz": fs/2}. beta is None when the usable segment is under 1024 samples or fewer than 16 bins survive the band mask.

The convention is the familiar one: β ≈ 0 is white, β ≈ −1 pink, β ≈ −2 brown. More negative means energy concentrated low, which the review associates with calmer stimuli.

Fixed by design

The 50 Hz lower edge and the log–log fit are fixed and documented rather than configurable. Slope is exquisitely sensitive to band choice, and a parameter that each lab tunes silently is not comparable across studies — the whole point of the guideline. Override the function explicitly if you must deviate, so the deviation is visible at the call site.

Short clips

The FFT window is rounded down to a power of two no larger than the signal. Before v0.2.0 it rounded up, which raised an exception on clips under about 1.4 s.