Opus for movie soundtrack?

[Mephesto]

Opus at its present state defeats AAC for speech and broadband audio while it sucks with tonal audio and because a movie soundtrack is a mix of all types in one I'm stuck on whether Opus might outclass AAC in this department.

I've always wished I could use 64 kb/s for a stereo movie track which HE-AAC achieves with great quality on 90% of it but if there's just one part of the movie with things like glass clinking, chimes, bells or triangles then HE-AAC fails horribly on these parts while sounding flawless with the rest. Because I accept no such compromise, I've always used 80-96 kb/s AAC.

I'm wondering if Opus might finally make that possible? It uses really low bitrates for speech parts and another algorithm for more complex ones so even though Opus does suck with tonal audio maybe the extra bits saved from the speech segments could make up for the loss in those parts? Has anyone done tests on this?

[Selur]

Have you tried using LC-AAC instead of HE-AAC, some people prefer LC-AAC for bit rates like 64kb/s and up,...

I rarely encode to opus and never did any extensive blind tests, here's a small test:
1. I took the audio from http://freemusicarchive.org/music/Aint_No_Love/FrostWire_Creative_Commons_Mixtape_Volu..._MIXTAPE_VOL_4
2. converted it to:

ffmpeg -y -threads 8 -v -10 -i "C:\Users\Selur\Desktop\Aint_No_Love_-_01_-_Shine.mp3" -ac 2 -ar 44100 -f sox - | sox --multi-threaded --ignore-length --temp "H:\Temp" --buffer 524288 -S -t sox - -b 16 -t wav - rate 48k | opusenc --bitrate 64 --comp 10 --framesize 20 --expect-loss 0 --max-delay 1000 --ignorelength --raw-bits 16 --raw-rate 48000 --raw-chan 2 - "H:\Output\test.opus"

Attached both the original mp3 and the reencoded opus file, assuming you use MPC-HC or similar players you should be able to playback and compare both.

Cu Selur

Ps.: I used 'opusenc opus-tools 0.1.7 (using libopus 1.0.3)'

[Mephesto]

No, quality of LC-AAC at 64 kb/s is unacceptable. The audio is dull and smeared, sounding a lot like 96 kb/s MP3. SBR perfectly compensates for this when the accuracy of the upper frequencies it replicates do not matter in the case of speech or casual audio but it does matter when the high frequency details are not uniform like instruments with high notes. It would be great if there was a way to selectively apply HE and LC to parts of a movie soundtrack but alas.

Your sample is a song not a movie soundtrack so it doesn't tell me anything. What I wanna know is if I encode a 90-minute film soundtrack at 64 kb/s where over half of it is casual audio and speech which would likely be encoded by the SILK algorithm and the more complex parts with CELT. Say the SILK parts are done at 32 kb/s and the CELT parts at 96. I'd like to know if the overall quality, at any given moment in the track would be better quality than HE-AAC.

I've done listening tests with Opus just not with anything more than a minute long.

[Mephesto]

Ok so I tested this out on my own since it seems no one tried this. I encoded the three-hour soundtrack of titanic to 64 kb/s opus and HE-AAC.

What I failed to consider is that even during parts where there's only chatter, it's usually accompanied by noticeable ambient noise from the ship, many times a lot louder when they're outside, and Opus is notably great with preserving this kind of broadband audio. So I'm not really sure if the SILK algorithm was summoned that many times at all or if it's good with speech sprinkled with background noise.

I also remembered Opus better preserves stereo separation than AAC (albeit with the expense of more distortion) so I decided to test this out with a part in the movie (first uploaded file) and subtract the channels. As you can hear, AAC preserved less channel information and upper freqs are distorted. Opus did a lot better in this depart, though I gotta say the first few seconds may rank as the weirdest goddamn artifact I ever heard.
"Caution: if you hear pissing sounds in your encoded audio, it must be Opus with a low bitrate!"
I swear, I actually thought my girlfriend forgot to close the bathroom door when I first listened to the Opus one. Wtf. Original sounds nothing like that.

Good news is Opus in general preserved the stereo a lot better than AAC. Now I understand why HE-AAC sounds so dull sometimes despite the upper shelf not being gone, because it ****s up the stereo. Durrrrr.

However, I'm disappointed how it handled the more tonal parts which I was hoping it WOULDN'T **** up so much on but it did. Check the second sample set. Opus made it sound wobbly, the thing we dreaded from the 128 kb/s MP3 days.

Also, does it really have to add noise where there isn't? Check third sample set. AAC perfectly retained the quality of this violin scene despite the spectrograph looking like shit, Opus didn't do too bad but why the hell does it have to add that gravelly noise or distortion in parts that have nothing of the kind? Unnecessary...

So to answer my own questions, I don't see evidence of any "intelligent reserving" of bits to parts that need it, though even bitrates like 96 with Opus still add noticeable distortion to tonal audio so we'll never know.

Overall, Opus did better than HE-AAC. It had better stereo and accurately retained far higher frequencies (including half of that inaudible 16khz beep present in all movies). I'm impressed.