Current mp3gain alternatives for reducing clipping?
Reply #3 – 2012-12-18 11:03:38
To test whether it's clipping or expectation bias (which we ALL suffer, hence the need for ABX testing and TOS #8 to ensure claims are verifiable on these forums): - you can decode the original full volume MP3 with clipping causing flattened peaks when there would have been overs then reduce its volume by, say 6.02 dB (6.0 is close enough) in Audacity or similar editor and save as a 16-bit or greater WAV (PCM) at 44100 Hz sampling rate - use mp3gain to apply -6.0 dB gain (which is actually -6.02 dB rounded) so they're matched for loudness and decode that (e.g. open it in Audacity and save it in the same lossless WAV format as the other sample. Use any ABX tools such as Foobar2000's ABX Comparator or PC-ABX or ABC/HR for Java (which you can use on a Mac) and choose the area to test and the number of trials (e.g. 8 to 10 trials). I admit I've thought things and written about them on these forums that an ABX before I hit send has shown to be inaudible (in fact one of these was clipping-related - a huge multi-sample clipped over picked up by mp3gain scan at the end of a Rachmaninov symphony, which I tested as above and found to be indistinguishable from the unclipped version amongst the loud crash of percussion and full orchestra. I'd convinced myself of a nasty noise in unblind testing but was unable to tell it from the clean sound or the lossless original scaled to the same reduced loudness once I didn't know which sample I was playing and trying to match with the known versions.) Some very old presets for encoders such as LAME used to include a --scale switch to reduce the likelihood of decode clipping but ABX has not demonstrated an audible difference even when clipping occurs. Another unlikely possibility is that your hardware (ADC) is non-linear near full scale, which would be a big surprise in this day and age, and while many high end amps are non-linear with respect to intersample overs (when viewed on an oscilloscope), they are very linear up to full scale and I'm not aware that any of them have ever been ABXed in music reproduction. You could record your soundcard's analogue output (e.g. via a 3.5mm stereo male-to-male cable into another computer) and ABX that against the output of a mp3gain-reduced signal to rule out any hardware problems. Personally, recognizing that the loudness of a CD relative to another is not intentional artistic choice and that there's no standard loudness, I apply Album Gain to all my music. I do not care about UNDO (and I keep the original lossless anyway so I could re-encode) and I usually apply the Album Gain in foobar2000's Convert... dialogue before it is sent to LAME to be encoded, so that when I scan it, Album Gain is usually very close to 0.00 dB. But feel free to enjoy your music as you like.