Odd sounds on MP3 export

[Moosebone]

When I export a session to MP3 the silence at the start and finish of the session is recorded over with odd digital-like sounds at very low volume. Something like old telephone signals or a bad '50's sci-fi movie. Anyone else getting this? It doesn't occur with WAV or AIFF mixdown exports. This seems like it's new from 10.1.1.10 or thereabouts.

[ryclark]

What bit rate and depth do you have set for the mp3 compression? Low bit rates can do this. Also the audio that you are compressing to mp3 may not be at optimal levels for doing extreme compression. How long is the 'silence' at the start and end of the audio and is it really absolute silence?

[Bob Howes]

Also, how many MP3 stages has your material gone through before your final version?

MP3 causes errors in your recordings and these are cumulative--eventually these errors hit a cliff edge and become objectionable.

Just as an example, it you took an MP3, opened it in Audition, it's converted to wave for editing/mixing then, if you save again to MP3, that's a minimum of 2 MP3 codings. If you went through several stages of MP3 coding during your editing--or if you got your source material for somebody who used MP3 all these codings add up.

As a matter of policy, you should keep material 100% in wave until  completely finished, then make an MP3 copy purely for distribution.

[Moosebone]

Thanks and yes, I'm definitely guilty of using MP3s with voice overs in WAV for podcasting. On experimentation doing the final mixdown in WAV and then converting the entire show to MP3 yields a more satisfying result.

For the record I was encoding at 256kbps CBR for the final mix down with a couple of seconds of pure silence at the start and finish - which is where the artifacts got inserted and why I could hear them.

Nonetheless, pre-Audition CC 2017 this wasn't occurring.

[ryclark]

Can you post a clip of the beginning or end of the bad .mp3 somewhere via Soundcloud or Dropbox so that we can hear the problem?

There have been several builds of Audition CC 2107 recently so which one are you using? And which version sounded OK previously?

[Bob Howes]

As far as I know all the 2017 versions use the same MP3 encoder. 

Two things to try:  first, if you have an old file that coded okay previously, try that on this version.  The way MP3 works, it can do very different things depending on your source material.

Second...a quick check that the MP3 settings have stayed the same (256kbps  CBR as you say) or did some of your preferences not transfer across.

[Moosebone]

Thanks for your replies - sorry for the delay in responding.

Here's a link to a sound file as raw output from Audition 10.1.1.11 multitrack mixdown on pure silence, MP3 256kpbs, CBR. You may want to turn up your volume. All the settings transferred over and as I say it seems unique to the MP3 encoding. It's also occurred via dynamic link from PremierePro on the return journey (in fact that's where I first noticed it). I'm wondering if it's a watermark of some kind. Interestingly the sounds don't survive media export from Premiere...

http://ak2ar.com/assets/multitrack-mp3-silence-mixdown.mp3

[SteveG_AudioMasters_ · Accepted solution]

It's not a watermark. Well, probably not in the sense you're thinking of, anyway. But, I do know what's happened. When you've saved your 'silence' you've started with a 16-bit int file, and the artifacts you've got are the sounds of lsb noise with an mp3 coding gain - you can tell this simply by the level they're occurring at, which is about -96dB (the 16-bit noise floor).

The reason that the noise is somewhat 'discrete' is that to code an mp3, the audio is split into 'critical' bands, and an evaluation is made as to where the inaudible bits are, and these are coded at a reduced bit depth (the masking process). On top of that, when it's silence, it's all about the lsb, and, especially if a file is dithered before saving, there's random noise at that level. Since a transform is done, you end up with a statistical removal, and this is the sound of the result. It's pretty much the same as overdoing NR - that results in a very similar noise. To sum it up, what you've got is, at its heart, coded dither!

Just to prove this to you, have another go. This time create a 32-bit file of silence (which is inherently not dithered), and convert that to an mp3. I have tested this, and I saved mine at 320k, although this should make no difference. Anyway, try it and see what you get then...

[Moosebone]

I'll be hornswoggled! That was it. Thanks Steve! Like you said, on experimentation the difference seems to be in using a 32-bit depth vs. the 256 or 320kbps format setting.