Can someone tell me if Itakura-saito distance calculated in this way gives the value in dB?
If no, then any opinion or suggestion?
dB computations or calculations are in addition or subtraction.
In fact, that is one of the two reasons going to logarithm in first place.
The equation mentioned is an integral with typical metrics (scalar not dB).
The integral is over d(omega) and then divided by 2pi, so the result is frequency, Hz.
i have never heard of the "Itakura-saito distance" before. i did know about the Log-Spectral distance which is in dB.
the form of this the integrand is (x-1) - log(x) and the first term is an approximation to log(x) for x close to 1. seems to me that this could be scaled by some factor like
20/log(10) = 8.685889638
or maybe half of that if the spectrum is power spectrum, and then the result might be able to be interpreted as "dB" in some manner.
i don't really get this metric. because it looks like it can have a negative value or positive. no square of distance nor abs value of distance.