Hector Martin

This is what happens when you transcode the same audio file 4 times with @ffmpeg AAC, vs. libfdk_aac, at 128kbps (the docs say it's better than libfdk_aac at that bitrate): https://mrcn.st/t/orig_ffmpeg_fdk_4x.wav … looping {1 bar lossless - 1 bar ffmpeg - 1 bar fdk}

This repeated transcoding is *not* a fabricated use case. In fact, I expect *5* or even *6* generations of lossy encoding for portions of an event I'm holding on Sunday (albeit the two gens I can control, at 320kbps, to minimize loss).

It does make it very clear that something is terribly wrong with ffmpeg AAC; I'm using it at 128kbps here as an example where *anyone* will be able to tell, so we don't have to argue over whether I can really hear the difference or not for more common 1 or 2 encode cases.

ffmpeg 4.2.2, source audio: https://mrcn.st/t/btc_bridge.wav … Command line: ffmpeg -i in.wav -c:a {aac or libfdk_aac} -b:a 128k out.mp4 (then decode to wav and repeat another 3 times)

And if anyone is wondering how you end up with 6 generations of lossy encoding, it goes like this: → original song (lossy fmt) → remixed (lossy fmt) → streamed by DJ to VJ for visuals → streamed by VJ to me for mixing → streamed to Twitch for broadcast → Twitch transcodes

I haven't checked that tbh, I just kind of assumed it did at least under some conditions, given that people can feed it weird settings. I'll be glad if it doesn't. Still up to 5 transcodes though.