>>815097
afaik foobar2000 uses r128 internally, and it's comfy
but it's still not perfect because simply adding silence to a record changes the result, while it shouldn't.
a sane way to measure "perceived loudness" would be like this:
1: completely ignore parts which have peaks below for example -60dB
2: compute RMS values of the record after equalizing with inverted equal loudness curve
3: for each channel, get mean of values above 80% percentile
4: ??? it depends on the number of channels, the final result must be somehow derived from the per-channel results.
points 1 and 3 is where (most) current implementations are lacking