MP3 VBR Transparency Rate
Reply #2 – 2009-05-01 13:30:31
... What I'm wondering is that, what is the transparency bitrate for VBR MP3 for most people in here? ... If you look at last fall's 128 kbps mp3 listening test (see the listening tests forum) you will find that all of the mp3 encoders tested yielded a perfect or close to perfect quality to the average of listeners on the average of samples. In the case of Lame 3.98.2 that was at a setting of -V5.7. So in a practical sense you're fine with a VBR setting yielding a bitrate of 128 kbps on average. Taking a closer look you'll find that what is valid for the average of samples isn't true for any sample. There were samples in the listening test that provided problems for the one or other encoder, and the quality in these cases was lower to essentially lower than what was the case in the average view. That's why many mp3 users allow for a security margin and use a higher bitrate. Quality improves with those hard-to-encode samples and can become perfect. Ratio of quality vs. bitrate increase is better in the moderate bitrate region, that's why according to last year's poll most Lame users prefer a Lame VBR setting around -V2 (which means an average bitrate of roughly 200 kbps or a bit lower). On the other hand many users don't have storage space problems and can allow for -V0 (roughly 240 kbps on average). In fact this group took second place in last year's poll. With such a setting you are very much on the safe side (though there are still samples not encoded perfectly). If 240 kbps is too much you can reduce average bitrate by using mp3packer afterwards (a lossless and very fast process) and/or restrict the extreme high frequency area to 17.5 or 16.7 or 16.0 kHz or in the variable way of the -Y option according to your likings. (You'll probably find that you don't miss anything beyond 16.0 kHz. -V0 --lowpass 16.0 and the mp3packer procedure yields an average bitrate around 215 kbps). All depends on how much you care about problematic samples. Probably it's wise not to do too much (there's a chance you personally never encounter one), but maybe it's also wise not to ignore them completely. It's up to you where to put your personal sweet spot. Maybe a rough overview in problem areas can help you decide how much to care about problems: - electronic music Some of the worst problems arise from this genre, especially when it's about sharp attacks of impulses artificially produced here. If you are a lover of this genre you should be problem-aware up to better not use mp3. - harpsichord music. Very serious issues for mp3 can arise from harpsichord music. With very high quality settings however quality of a good mp3 encoder should be very acceptable at least. - long-stretched tones of whatever origin Long-stretched tones can produce distortions and/or a tremolo-like effect when using bitrates too low. - Voices Voices can be encoded with subtle distortions when using a bitrate too low. Usually voices are encoded fine even at low bitrate, but it's not always like this. To me french woman singers are pretty prone to this issue, but last night when I encoded african music at moderate bitrate I ran upon this issue also with male african voices. I guess we're very sensitive to voice, so errors at the same subtle error level are recognized easier than when originating from instruments.