centroid is calculated alwais using hanning window

[alalbiol]

If spectral bandwidth is requested with other window than hamming, the spectral centroid should be calculated using that window.

The following code in the function should be fixed:
# centroid or center?
if centroid is None:
centroid = spectral_centroid(
y=y, sr=sr, S=S, n_fft=n_fft, hop_length=hop_length, freq=freq
)

should be:
# centroid or center?
if centroid is None:
centroid = spectral_centroid(
y=y, sr=sr, S=S, n_fft=n_fft, hop_length=hop_length, freq=freq, window = window
)

[alalbiol]

There is also a related issue with this. spectral centroid always recomputes the spectrogram even if it is provided as parameter. The normal behaviour should be not to recalculate spectrogram S if provided

[bmcfee]

Thanks for noting this. I believe the code actually is correct here, but it's not obvious upon a first read through. Here's what's happening:

• In these feature functions, either the signal (y) or spectrogram (S) can be provided as input. When S is not provided, it is computed by calling the _spectrogram function.
• If _spectrogram is called with S already provided, it behaves as a no-op and returns whatever value of S was provided initially. So your second point is not actually correct: the spectrogram is not recalculated.
• When spectral_bandwidth first computes the spectrogram (if needed), all parameters are passed in correctly (including the window function). This spectrogram is then handed over to spectral_centroid later on, and no recalculation takes place: the correct window is used here.

It's true that the code would be more readable if all parameters were correctly passed through to spectral_centroid, but I don't think it would actually affect the behavior.