Looking for the most efficient (easiest to decode) format for iPod 5

[speachy]

Those opus benchmarks are a reminder that I really need to optimize the non-power-of-2 FFT. Been on my to-do list for 12 years now.

We last resynced our opus snapshot with upstream in April 2019; it would probably be worth doing that again before putting any additional effort into local optimizations, though I can't imagine they've given any thought to any of the CPU architectures we currently care about (ie "Classic" ARM, m68k, and even MIPS..)

(That probably goes for most of our codecs, fwiw...)

[K4sum1]

Those opus benchmarks are a reminder that I really need to optimize the non-power-of-2 FFT. Been on my to-do list for 12 years now.

I would be curious how a well optimized Opus decoder would fare on the iPod. Although even if it's a substantial leap maybe I'd still want to go with ogg because VLC seems to have a weird delay playing Opus files for me.

I was also curious how Vorbis encoders fare, and I also discovered a tool called OptiVorbis which (slightly) decreases file size without any impact on quality (the spectrals are the same) So I ran some tests using my own custom files. I also wanted to see how fast wav would decode.

aotuv_128.ogg
175906 of 175906
Decode time - 49.19s
File duration - 175.90s
357.59% realtime
22.37MHz needed for realtime

aotuv_144.ogg
175906 of 175906
Decode time - 49.85s
File duration - 175.90s
352.85% realtime
22.67MHz needed for realtime

aotuv_160.ogg
175906 of 175906
Decode time - 51.03s
File duration - 175.90s
344.69% realtime
23.20MHz needed for realtime

autovopti_128.ogg
175906 of 175906
Decode time - 49.30s
File duration - 175.90s
356.79% realtime
22.42MHz needed for realtime

autovopti_144.ogg
175906 of 175906
Decode time - 49.89s
File duration - 175.90s
352.57% realtime
22.69MHz needed for realtime

autovopti_160.ogg
175906 of 175906
Decode time - 51.23s
File duration - 175.90s
343.35% realtime
23.29MHz needed for realtime

vorbis_128.ogg
175906 of 175906
Decode time - 49.11s
File duration - 175.90s
358.17% realtime
22.33MHz needed for realtime

vorbis_144.ogg
175906 of 175906
Decode time - 49.70s
File duration - 175.90s
353.92% realtime
22.60MHz needed for realtime

vorbis_160.ogg
175906 of 175906
Decode time - 50.70s
File duration - 175.90s
346.94% realtime
23.05MHz needed for realtime

vorbisopti_128.ogg
175906 of 175906
Decode time - 49.17s
File duration - 175.90s
357.73% realtime
22.36MHz needed for realtime

vorbisopti_144.ogg
175906 of 175906
Decode time - 49.81s
File duration - 175.90s
353.14% realtime
22.65MHz needed for realtime

vorbisopti_160.ogg
175906 of 175906
Decode time - 50.86s
File duration - 175.90s
345.85% realtime
23.13MHz needed for realtime

wav_16-44.wav
175906 of 175906
Decode time - 3.97s
File duration - 175.90s
4430.73% realtime
1.80MHz needed for realtime

aoTuV is barely slightly slower to decode than libvorbis, and files run through OptiVorbis are barely slightly slower as well, but outside of being barely measurably slower here, it's way too small to be noticeable in normal use.

People say aoTuV is better, but the spectrals from libvorbis look better to me. A lot of the discussion about aoTuV being better is ancient too. However it still seems aoTuV may still have some improvements, but I'm not sure how to quantify them. I know I could do some like blind test, but I feel like I wouldn't be able to notice the difference so I don't even want to try and have been trying to find some scientific or test collection that isn't ancient to point me towards what to use, but that just doesn't seem to exist or I can't find it.

Also the wav result is insane.