Audio Levels for YouTube-Only Viewing

[Scott Wilkinson]

Our production house has many individuals who come from a broadcast television background (which is great). But---for better or worse---the vast majority of our video production is viewed ONLY on YouTube---there is ZERO chance that it will ever be repurposed for broadcast television.

As a result, audio levels for much of the work my people produce are---in my opinion---far too low for YouTube playback. My producers generally master with peaks anywhere from -12 to -20dB.

In my experience---based on watching thousands of videos on YouTube (and listening to them on a pair of good-quality near-field monitors)---broadcast audio level standards are just too low for an all-computer-based stream with no broadcast elements whatsoever.

On a more practical (and irritating) level, I notice that our vids are consistently quieter/harder to hear (without cranking up the volume on my MacBook Pro) than other well-produced YouTube videos.

I do understand that EQ and compression have *something* to do with perceived loudness on the YouTube side...but I still think for a YouTube-only workflow, aiming for peaks at somewhere around -3dB is much more realistic.

As a side note...when we produced our national television spot (aired during football games on the Fox and ESPN networks)...the networks mandated that our audio be mastered at -20dB.

I was dubious of this...and was flat-out pi$$ed-off when, while watching our spot on national television (on a good-quality flat-panel HD television)...the perceived loudness of our spot's soundtrack was DRAMATICALLY lower than the Toyota commercial that aired just before it. The difference was FAR greater than what could be accounted for by compression. It was blatantly obvious to me that the Toyota commercial had been mastered at -12dB or higher...because it BLASTED out of the set. (Ours sounded like a half-hearted whimper by comparison.)

Thoughts?
Scott

[Robert Benda]

For our work (weddings), I tend to try and peak at about -6dB, which gives me a little bit of room for when I have no choice but you use non-ideal audio and pump it up - I'm using a lot of ambient audio - or leaves a little head room for when there is a sudden spike like a yell, cheering, or applause.

For soundtracking with voice over, I find the music needs to peak at no more than -18dB, and a little lower is better is the music has vocals.

[Rick Reineke]

Same as Scott, I run web based program audio peaks around -3dBFS as well. (using 'properly adjusted' compression/peak limiting) Higher peaks could also result in distortion when YT / Vim re-encodes it.
FWIW, I encode video material destined for the web using HandBrake, from a high-quality intermediary. (DNxHD in my case). This seems maintain video/audio pretty well even when it's re-encoded.

"Our spot's soundtrack was DRAMATICALLY lower than the Toyota commercial"
The 'CALM act' (ATSC A85 recommended loudness) was designed to counter the loud commercial syndrome.
The local station is likely not in compliance with the current loudness standard. (and subject to FCC scrutiny, fines, ect.) The broadcaster 'should have' rejected the commercial or at least re-adjusted the audio level.

[Gary Nattrass]

I agree -6 or -3 works better for on-line work!

I usually do a broadcast level master pro res and then do an mpeg4 for you tube with +6db added!

[Scott Wilkinson]

Thanks for the replies---glad to see at least a couple of you don't think my suggestion of -3dB is crazy. :-) (Some of my broadcast-oriented producers do.)

@Rick Reineke: I'm familiar with the CALM Act, and you're right about that. In the case I mentioned above, I'm pretty sure that was a national Toyota corporation commercial (not a spot for a local dealer). And this was on FOX television.

It ticked me off because it seemed like a blatant case of..."We'll let the top-dollar-paying corporate guys pump it up to -6dB...but be sure the freebie university PSA guys stay down low at -20dB." (There is no way that Toyota spot complied with the CALM Act.)

I could be wrong, but...

I think this issue (of audio levels for online-only delivery) is just another of many examples where traditional broadcast-oriented workflows no longer apply. The other issue, of course, is video specs...where traditional IRE levels are basically meaningless when you live and work in all-digital, direct-to-YouTube workflow with no broadcast component at all.

Scott

[Richard Crowley]

It is not completely clear why we should pick such a low value as -3? We aren't talking about mixing live feeds where there may be unpredictable peaks. We are talking about fully POST production where we know exactly what (and where) the peaks(s) are. What is the argument against uniformly normalizing the entire track to -0.5dB or even -0.1dB after everything is mixed, mastered, etc.

This is a completely different paradigm from legacy broadcast standards. By normalizing to some arbitrary level below 0dBFS, we are deliberately discarding perfectly good available dynamic range in the delivery system.

[Seth Bloombaum]

Originally Posted by Richard Crowley

It is not completely clear why we should pick such a low value as -3?... What is the argument against uniformly normalizing the entire track to -0.5dB or even -0.1dB after everything is mixed, mastered, etc...

Quite right, at -3 or -6 we're leaving something on the table. On the other hand, normalizing to these levels puts your clip in line with other loud clips, not significantly above.

My yardstick may be a little skewed, because I routinely use the excellent "Loudness" filter in Izotope Ozone for dialog-based web content. Perhaps my -3 is equivalent to someone else's -0.1.

[Tom Morrow]

Peaks are a pretty inexact way to determine loudness; as you are certainly aware the amount of dynamic range in the subject material, recording and mixing paths can give you vastly differing loudness for a given peak level. I am no expert but have researched this problem and decided the best approach for me is to simply compare the loudness by ear with a reference that is chosen to have good levels for the medium.

Google LKFS for more info than you can deal with about new standards that do a good job of quantifying loudness in a meaningful way.

To cut to the chase, I think it's entirely valid to let peaks go to full scale for web mastering, but depending on the dynamic range sometimes they might be below full scale and sometimes you might need compression to master signals with above full scale.

As to commercials being louder than TV programs, it's no secret that is by design not accident.

[Richard Crowley]

Originally Posted by Tom Morrow

Peaks are a pretty inexact way to determine loudness;

Of course. But subjective "loudness" and objective "dynamic range" are completely different topics.

To cut to the chase, I think it's entirely valid to let peaks go to full scale for web mastering, but depending on the dynamic range sometimes they might be below full scale and sometimes you might need compression to master signals with above full scale.

I don't understand that? "Dumb" normalization will take you safely up to exactly 0dBFS. Normalization is a mathematical manipulation (simple addition). Remember this is AFTER we have done whatever compression, limiting, clipping, etc. etc. "loudness processing".

OTOH,I have seen some discussion about normalizing to 0dBFS causing clipping problems with some playback systems. Because of the nature of how MP3 (or similar compression algorithms) re-construct the audio waveform "overshooting" full-scale in some cases. Which is why some of us normalize to -0.5dbFS or -0.1dBFS if delivery is going to be data-compressed (MP3, et.al.)

[Rick Reineke]

My M.O. of -3dBFS (highest peak), is without excessive compression and volume maximizing of the master PCM file. The absolute peak of -3dBFS (in my case) leaves adequate headroom prior to encode/re-encode process. Peaks of around -1dBFS and higher would almost certainly be clipped in and after the encoding process.
FWIW, measuring one my video clips audio on Vimeo http://vimeo.com/31228677#t=0s yields the following statistics:
Minimum sample position (Time): 00:02:04.125
Minimum sample value (dB): -1.335
Maximum sample position (Time): -00:02:56.327
Maximum sample value (dB): -0.659
RMS level (dB): -19.960
Average value (dB) : -Inf.
Zero crossings (Hz): 1,391.16
Maximum true peak sample position (Time): 00:02:56.327
Maximum true peak sample value (dB): -0.656
Maximum filtered true peak sample position (Time): 00:00:29.571
Maximum filtered true peak sample value (dB): -0.530
Integrated Loudness (LUFS): -19.77
Loudness Range (LU): 4.80
Maximum True Peak Loudness (dBTP: -0.66
Maximum Short-Term Loudness (LUFS): -16.10
Maximum Momentary Loudness (LUFS): -12.32

[Josh Bass]

As others have said, -6 for regular dialogue peaks for me. Have a regular corporate client who we make videos for with mass internal distribution meant only for intranet-web viewing. He's told me that when he mastered to -12 for peaks in the past he's heard many complaints of the audio being too low.

"Loudness" is hard to measure with meters, has sometimes little to do with peak levels, and for me is best judged by ear (e.g. If i was trying to match the loudness of one thing to another i would a/b them until they sounded equal, and THEN work on taming any errant peaks with compression/limiting/automation. You can have a single peak that spikes to -3 but if everything else is down in the -12s itll still sound "low". You have to bring those things more in line with each other.