jcjr--Youtube uploads

[jancivil]

I don't know anything about 'true metering'. I did a couple of tests, one of mine which was getting a penalty 100 = 71%, got 4% less of a penalty reducing two peaks (no clipping, mind) by like 1dB. I don't really care that much but I am re-rendering this one doubling the video resolution to 3840x2160 to see if the Opus code happens. This render took over 4 times longer than 1920x1080.

[jancivil]

BTW, Da Vinci Resolve costs no money, except in the fancy version which provides for extras in collaboration compatibility which I understand nothing of, and I imagine some extra hardware integration features. I got hold of it strictly to get wav or aif audio in videos. Cubendo used to do Replace Audio in Video but Steinberg killed the feature a couple of yrs ago, and it's still dead.

Well, that got rid of the normalization penalty anyway.

[pekbro]

I think you can do the replace audio thing in Sonar (now the free Cakewalk). That's about
all I used to use Sonar for, was its ability to work with video. No doubt it still remains.
Sadly, it wont help anyone on a mac, like jan...

Another thing, if anyone happens to have an adobe cc account (even the basic photo)
you can use media encoder, which is pretty comprehensive.

Anyway, interesting topic. I have pretty much decided to spring for vimeo so as to avoid
any hassle with youtube myself. I'd like to avoid problematic rules, trolls etc, even if I gotta pay.

[JCJR]

Comparison AAC vs OPUS

AAC : youtube.com/watch?v=9iQKHI8W2yg
OPUS : youtube.com/watch?v=ByBFkRL07FE

I can see the difference without problem

Opus win !

Thanks clothem. I could see a bigger visual difference in your previous no-audio picture example that you linked. I looked at your above-linked two music video files on the old kindle fire and my 34" QHD "main" monitor of my studio 3 monitor setup. Dunno if I could reliably do a double-blind discrimination. Might later check it on other computers.

Blame it on my eyes. Dunno if it is the years of glaucoma or the years of poisoning from the glaucoma meds. I still measure 20/20 but contrast ain't what it used to be.

Uploaded this latest one 1440p, 24bit 44.1k samplerate WAV via Virtual Dub, but did not try that video encode setting trick you recommended yet. Maybe next vid. Immediately after upload "stats for nerds" showed mp4a.40.2(18), 96kbps, 48k samplerate. After a few minutes, it re-encoded and showed mp4a.40.2(140), 128kbps, 44.1k samplerate.

And now awhile later after embedding here, YT seems to have got around to re-encoding to Opus(251) as in your example, 160kbps, 48k samplerate.

https://www.youtube.com/watch?v=fHeVUuLxJso

[jancivil]

YT re-encoded mine as well, re-penalizing me 25% but it's Opus now.

[jancivil]

Ha! I just noticed Nuendo has this option in Export Audio: Normalize to Integrated Loudness. Defaults -23 to -1.

[JCJR]

YT re-encoded mine as well, re-penalizing me 25% but it's Opus now.

Hi jancivil

I don't know if it matters so much if they might normalize a little bit-- 25 percent is only -2.5 dB cut. If you have uploaded uncompressed audio, I read that google/YT keeps the original uploaded file "forever". So if they "do it right" then they would normalize the uncompressed original audio and then re-encode, which ought to be about as transparent as an uploader trying pre-empt it ahead of time.

Of course they might not be "doing it right" and the normalization might be more damaging to folks who upload pre-encoded already-lossy audio. Or maybe they have a way of merely sending a "master volume" instruction embedded in the stream, which ought to be quite innocuous. Just wild unsubstantiated guessing as usual.

You maybe don't care about nerdly discussion of true peaks so I try to keep it brief, though being a nerdy fella it is like trying to keep a pig out of mud.

Before maybe a decade ago, most peak limiters would limit based on the sample values encountered, and Limit Ceiling settings for CD mastering were often in the ballpark of -0.1 to -0.3 dB. But when the audio samples are reconstructed into a continuous analog signal smooth bandlimited curves get drawn thru all the sample points. So except for the unlikely circumstance of a sample that just happens to lie at the very top of a wave peak, that "smooth connect the dots" will go higher than the samples lying on each side of the "true peak" location in time, somewhere between the two samples.

Depending on the nature of the audio, these "true peaks" can be a dB or louder than the actual sample values. Some DACs have plenty of headroom and never clip on such mild overs but other DACs might clip.

With data-compressed audio or samplerate converted audio it potentially gets worse. Samplerate conversion "done right" tries to act like a perfect DAC reconstruction filter, and then resamples that "perfect reconstruction" at different time locations. So if the sample peaks go as high as -0.1 dB and the "true peaks" inbetween the samples are often bigger than 0 dB, then after resampling you have good odds of having resamples at places higher than 0 dB, causing multiple clips in the file.

There is also something called the Gibbs Effect, which causes peak levels to paradoxically get bigger if you lowpass filter an impulsive audio stream. Nowadays just about all the TV and internet radio and whatnot is data compressed. Even if they don't samplerate convert it, they typically lowpass the audio either a little or a lot (depending on target bitrate) before doing the encode. So if you have an audio file with sample peaks "real close to 0 dB" and compress it to 128kbps or less, there is enough lowpassing going on to add possibly some serious peak clipping during the encode process.

So that is why the last decade's worth of limiters tend to read true peaks, and why recommendations have tended toward a max true peak level of about -1 dB. That seems a lot of wasted headroom compared to old thinking, but samplerate conversion and data compression will tend to raise the peak levels so the original audio needs some safety room right up at the top.

Apologies belaboring stuff not so interesting. Uploaded yet another moldy song with cobwebs hanging off, maybe even less interesting.

https://youtu.be/RQD9_d6ujtg

[JCJR]

I think you can do the replace audio thing in Sonar (now the free Cakewalk). That's about
all I used to use Sonar for, was its ability to work with video. No doubt it still remains.
Sadly, it wont help anyone on a mac, like jan...

Thanks pekbro I ought to take a look at the free Cakewalk sometime. For PC, that VirtualDub is as fast/painless imaginable for static picture + music but it might be painful trying to use it to actually edit a home movie or whatnot. But assembly of the merged avi is just about instant, just open the BMP file, open the audio file, open Frame Rate menu item and tell it to set video duration to audio duration. Then save as AVI, which is near-instant, done about as quick as I name the file and hit OK button. No encode happening at all, just writing the two chunks unmodified to a single AVI container file. And it is a very small file to upload, or as small as its gonna get if uploading stereo uncompressed WAVEX 24 bit audio.

A few years after retirement I thought back to my most pleasant DAW usage and it was Opcode Studio Vision and MOTU Digital Performer but I quit updating DP in early 2000's because too busy programming to record. Was thinking about getting DP again, but it has that dang pace CP. I can't stand CP any more complicated than a no-call-home serial number. So I tried reaper which I generally like, but wish the midi track operations were more elegant. Haven't found anything it can't do but recording MIDI is somewhat slow and bass-ackwards to what I'd prefer. Cakewalk/Sonar started out pretty slick for MIDI sequencing and then later Sonar versions, to my taste, seemed like they were making the operations more bass-ackwards and not less so, but haven't seen its current state.

Heck this old thang was done in the ESQ-1 8 track sequencer, driving various synths, analog mixed in real time to Sony PCM501 / Beta video tape 16 bit digital audio.

https://youtu.be/rE3tkDhqDZc

[JCJR]

More than half finished uploading creaky old instrumentals with cobwebs hanging off. Space Opera music.

https://youtu.be/BBAuszrXLbE

[JCJR]

More Space Opera.

Definitely a first-world problem but I have a gripe with some modern synth music homage to classic space opera. The stuff was written in the 1920's to 1940's when a theremin was a big deal, and big band music was the thing and even banjos and torch songs hadn't got uncool. So homage to steam punk/diesel punk that sounds the same as William Gibson's (already obsolete) version of the future is too much anachronism. Its supposed to be the way the future was supposed to be, not the way it actually turned out.

https://www.youtube.com/watch?v=2syiHRY3vzk

[jancivil]

I read that google/YT keeps the original uploaded file "forever".

They do. There is a google feature "takout" where you may download all your youtube, and the original audio is in these files.

[JCJR]

I read that google/YT keeps the original uploaded file "forever".

They do. There is a google feature "takout" where you may download all your youtube, and the original audio is in these files.

Thanks for letting me know, jancivil. Good feature!

https://www.youtube.com/watch?v=NuYamKpADno

[JCJR]

After this one only 3 ancient moldy instrumentals to go but maybe I'll keep it up for a dozen or so equally creaky old vocal numbers.

https://www.youtube.com/watch?v=EwNgiEyLmrE

[JCJR]

An udder 1.

https://www.youtube.com/watch?v=vNQwQ6Xo1PQ

[JCJR]

After this only one more ancient instrumental to upload. Have been working on newer stuff (whoopee or whatever) but am slow and imprecise. Maybe get more new stuff done eventually.
https://www.youtube.com/watch?v=Az6yxp7COY0