Posted by reply on 12/19/06 02:46

On Sun, 17 Dec 2006 22:54:29 -0500, "JJ" <up@yours.spammers> wrote:
|
|Well, I base my statement on the fact that I imported both the original file
|and the converted file into Audacity and the audio in the latter is visibly
|shifted forward in time. It was here that I made my measurement to determine
|the time spans I mentioned in my original post.

Now that I think I can see what you are referring to, I can only add a few
comments -- none of which will really help you in your quest.

It appears that almost every mp3 encoder I've tried since you posted throws a
bit of apparent silence ranging up to a frame length or two at the start. I
say apparent silence, because if you expand the vertical scale you may see a
couple of very low level audio.

But you shouldn't interpret that silence as a delay, for the waveforms aren't
the same. Look within the waveforms for some peak or cluster of peaks, and
see if that is time shifted, back or forward. Those are what you'll be
hearing.

On the small sample set I tested using Audacity & straight Lame (same result
whether I used command line Lame or Razor Lame), Audacity showed the smallest
startup time and also the smallest shift in the first high peak -- Audacity
mp3 actually peaked slightly before the straight wave. Of course this implies
a certain amount of trust in my waveform editor, Cool Edit and in the concept
of what constitutes a delay when creating a PCM file from an encoded version
of the audio.

I have to admit, I've never upgraded the Lame from 3.92 to 3.97, but I doubt
this is changing things.

Synthetic sounds, whether it be synthetic speech starting with LPC in the
early 70s or more modern audio compression such as mp3, isn't really meant to
match the waveforms, only the audio effect.

As an experiment, I used mp3trim to automatically trim silence from the start
of the mp3 files. It took one frame (.26 seconds) from the straight Lame
generated one, but left the Audacity one alone. In both cases the files were
left with a very very tiny blip in the first frame. Now the straight Lame
file peaks were shifted to the left of the original, by more than the Audacity
ones. By taking 2 frames from the straight Lame one or just one from the
Audacity one, the mp3s looked as though they started on time, but detailed
analysis showed that the peaks were too early if you're trying to match up
waveforms.

In conclusion, I don't think you can achieve the end effect you want without
some perceptual tests. mp3trim can be a good tool to trim bits off the end as
well as normalize the mp3s, but I don't think it will give you all you want.
Looking at the apparent delay at the startup may not give you what you want.
However, it might add to your fun in your research.

[Back to original message]