On 29/03/13 3:05 AM, Jostein Chr. Andersen wrote:
On 03/28/2013 04:43 PM, Raffaele Morelli wrote:
...
best result: lame --preset insane <infile>
The "insane" parameter name led me to understand that Audacity do use
Lame as well. I use the insane parameter when exporting to MP3 from
Audacity.In that case, I might not be satisfied with Lame at all.
But If "lame --preset insane" is the best thing in the Linux audio
world, then I might have to rethink how my mixes should be in order to
avoid the problems. One thing I have noticed is that esses (mostly when
accentuated from compression), dithering (or not) and tape saturation
affects the final audio result of a MP3 file.
LAME isn't just the best thing in the Linux audio world, it's about the
best MP3 encoder out there, full-stop; there's no way you're going to
get a better-sounding MP3. Even if you can hear differences, it's
incredibly unlikely that anyone else, who doesn't have your familiarity
with the source material, is going to be able to hear any.
In fact, I'd honestly be surprised if you really are hearing a
difference, or hearing the differences you think you're hearing. Doing a
proper ABX test, using a tool like Squishyball, could be very
instructive here -- you might find that you have a lot more trouble
picking the difference between the MP3 and the WAV than you think you would.
If you do want to do an ABX test, use LAME to decode the MP3 back to
WAV, and then supply that WAV to Squishyball, along with the original
WAV from before you encoded it:
lame --decode foo.mp3 foo.wav
The MP3 format adds a delay to the start of files during the encoding
process, and using LAME to decode it in this way will eliminate that
delay on decoding (assuming the original file was encoded with LAME,
too). If you don't eliminate that delay, then the results of the test
would be invalid, since you might end up noticing the delay, rather than
any real differences between the files.
In fact, I'd recommend this to anyone who's never done it before --
modern encoders are remarkable things, and there's no better way of
demonstrating this than with an ABX test.
Thanks
Leigh