Regarding Custom Mixes (particularly related to Video Games)

I’ve been curious as to how most people like Sonic, DAK, !!!!! and countless others do their mixes with (sometimes, lossy) content. I love compiling custom mixes from video games ripped by others on FFShrine. I do have a question in regards to mixing layers/pieces with very low bitrates together.

Ex: I have files from a game that averages on 220 VBR in OGG Vorbis. The songs are all in layers of 4 or 5 parts of the same length that play one after another, or in some instances, they all mix together in-game.

When mixing a custom soundtrack with lossy source files, is it better to keep the final edited output in FLAC or in MP3 320kbps (or even 256kbps, which is closer to the bitrate of the files mentioned earlier)?

Would love to hear everyone’s suggestions. Cheers

is it better to keep the final edited output in FLAC

Yes. If you convert to any lossy codec, you get loss of data/quality.

How much this is? You would have to make the versions you want and sit through a vigerous blind ABX test.

If you start a project, save your edits in lossless. This way, later on, if you change anything, it won’t be further degraded.
So don’t open ogg, edit, then save in ogg for later editing.

If you’re going to convert to lossy formats, try use the highest bit depth you can.
Convert to 32-floating point WAV. I doubt most encoders actually accept 64-floating point WAV.
32-floating should be good enough. 64fp will give the most accuracy, but do that only if you have the space for it. And time.
This will reduce any errors and add more precision when transcoding to lossy format. Just for rounding errors, though.
Samples that would do better with floating point math wouldn’t do well with cut-off integers.

For lossless, you can dither down to 16bit. No need for 24bit. Only for editing.

———- Post added at 10:32 PM ———- Previous post was at 10:21 PM ———-

Others will say resample your audio to 96kHz or higher for accuracy.

But I caution to really look into your software first. Some resamplers will cause aliasing and reverb artefacts when upscaled. You have to check your settings.
"Highest quality" should be good to use. It will be slow, but that’s the price you pay. Fastest resampling gives the worst results ever.
You’ll want to use the same settings to downscale back to 44.1.

For dithering, in lossless, you’ll want to check for that, too.
Some programs have a lot of options for the different algorithms. Some work best on different types of music. This no one knows.
Triangular-noise is most common.
eac3to will use TPDF (Triangular Probability Density Function).
Weiss Saracon has up to 4 options. Triangular, POW-r functions and "zero" or truncate. Some will just round off. Which I think is the same as truncate/zero dithers.

Further reading:
https://en.wikipedia.org/wiki/Dither
http://wiki.audacityteam.org/wiki/Dither

It’s the last thing you do. Some people who have no idea what they’re doing, "thinks" it gives the audio a unique quality or something artistic.
This has yet to be proven by actual humans.

Adobe Audition automatically processes at floating 32bit.