I have a basic understanding of dithering and why it’s used, and have been wondering what happens with repeated applications of dithering.
I so in my case, I have a 320 kbps CBR MP3 file that I imported into Audacity and made some edits (duplicated a section, fade in/out, envelope tweaks). I then exported it to FLAC with 24 bit depth with the default shaped dithering.
If i then re-import this FLAC and export it again to 16 bit depth FLAC or MP3, how much dithering will there be? Will the accumulated dithering be more than if I had just gone straight from an audacity project to 16 bit depth FLAC or MP3?
Dithering is added to a show when down converting (32 float > 16 bit) to keep errors from lining up and becoming audible. I suspect two passes through MP3 are going to do far more damage than anything Audacity is likely to do.
We very strongly recommend avoiding MP3 always, or saving it for the last step just before preparing a work for your iPod. Never as a step in the middle of production.
Yes, ACX requires AudioBook submission in MP3, but they have minimum quality requirements and we recommend exceeding them. By all means, produce the show and save your show archives as WAV.
It’s possible to export work and go through the conversion without dither. Dither is a preference setting.
It’s possible to export WAV at 32 Float, and it’s possible to Save an Audacity Project which doesn’t downconvert. All sorts of options are available.
It’s recommended every time you downconvert from 32-floating to 16 bit. That’s the step which creates the possibility of conversion artifacts. Upconverting is no trouble. You’re converting to a much higher accuracy format. But coming back down you can have a sound that wanders between two perfectly valid, well-behaved audio values in 32, but switches between two 16-bit values such that the wandering speed turns into sound. Suddenly you have sound in the show that didn’t used to be there.
It’s best to break up the show with a very tiny injection of random sound so at no time will anything ever be coherent enough to create damage. An early Audacity version used to choose tiny but audible dither values. That’s been fixed.
You could create an Audacity Project and downconvert to each of the products and deliverables as needed. As long as you stay in a Project, there is no dither.
MP3 is a time bomb. If you start with a 128 MP3 and cut it down to another 128, you don’t have 128 any more. You can have as low as half that. 60 is the quality limit for stereo MP3, so you can get into trouble pretty quickly with multiple compression passes. 128 used to be the default Audacity Export.
We have a poster who uses download MP3s to create a broadcast radio show for a station in rural North-East US. He is required to deliver MP3 and it’s almost impossible for the station to create the podcast version of the show. Too many MP3s.
I’ve been known to create super high quality MP3s to keep my lower quality original from getting much worse. The first time somebody does production and makes a new MP3, they are going to get the combined quality of the first and third MP3s, not 320.
I’m leaving the issue of MP3 out of this, because dither is only really relevant when dealing with high quality lossless formats.
Dither is added to prevent harmonic distortion from “quantization errors” (rounding errors when reducing the bit-depth).
The amount of dither applied when exporting as 24-bit is insignificant - it’s hundreds of times lower than 16-bit dither and will have no effect on 16-bit rounding.