correcting audio

[Chris Barnard]

HI all,

Using ffmpeg to create ac3 for my DVDs. I know with other (not free) programs there is a 'normalise audio' function that I imagine corrects the audio drops offs that happens in converting to Ac3. Don't want to pay for it.

The problem is evident when comparing live audio, specifically speaking persons, to music laid over. The live audio suffers a MAJOR drop off, while the laid over msuic suffers very little, leaving the whole thing out of whack. Volume-wise.

Just pushing up the volume of live audio in the project seems the most sensible solution.

Was wondering if anyone had any experience with this...?

[GrassValley_BH]

Volume adjustment is a tricky thing.

Normalizing audio in the general sense usually only does peak normalization.
That means it looks for the highest level, then raises the audio volume to the difference between that and the desired level.

For example, if a clip contains speaking, then sudden loud applause at -10 dB, and we normalize to -6 dB, then the audio of the clip gets raised (uniformly) by 4 dB. This doesn' make the applause any softer, it's just like turning the volume up.

Since it's just like raising the volume, if the speaking was soft to begin with, but the applause is really loud, then either you'll have the volume low and not hear the speaking (but not blow out your eardrums at the applause), or have the volume high, hear the speaking but blow out your eardrums at the applause.

The other trouble with peak normalization is that it take all the audio as a unit. This means if you have clips with low-mid-high-low levels and peak normalize the track, you'd end up with mid-high-max-mid levels at best.

So then you'd try peak normalizing each clip. Fine. The problem there is that each clip's peak level is now the same, but still doesn't help the "average" volume level which is what we interpret as whether the audio is loud or soft. Back to the speaking softly-loud applause clip, most listeners would consider the clip's audio "low" because the focus of interest is the speaking, not the applause. Unfortunately the computer has no subjective determination of what's "interesting" or not...

Which leads to RMS normalization, which attempts to give more of a "averaged" volume level. Oversimplified, RMS normalization "ignores" the peaks to a degree, giving a better "average volume" increase.

In the case of music, where there's loud hits or drum beats and such that might "distract" standard normalization, they use a "compressor." This is different from Zip and data compression - this is audio compression.

Essentially you can think of it like "squashing" the top and bottom. The peaks are "crushed" (it's an actual change) a bit, so you can raise the overall volume a bit higher. Compressors are often used in conjunction with limiters.

So there... Most audio software has compressor/limiter/normalize function. Compressors have some settings which I won't get into in detail, but essentially it controls when and how swiftly the peak-stoppage occurs.

[Chris Barnard]

Thanks Brandon - that helps a lot.

I'm actualy working on a project in Premiere 6.5 at the moment, and I've lowered the audio gain on the louder stuff and raised it on the quieter speaking parts. I'll see how that comes out.

If you lower the audio gain on some noisy music, are you saying that beause of the difference in the peaks and the troughs, the processing will still result in a louder sound (because of the greater variation in sound in a noisy audio clip) than a speaking clip, even if one has adjusted their respective audio levels so that they sound the same volume in the timeline - before processing the final project to a rendered clip.

How well do volume normalising engines handle this problem? Presumably not all programs are created equal?

Thanks for the help!

[STORMDAVE]

As Brandon said, sweetening audio is not as simple as hitting a button and "normalizing" it to 99% for example. You have to go to each source and normalize it.

My work flow is that whenever I do high end documentaries, I usually export each clip of audio separately, then render a final project into Canopus DV for example, import that into Audition 2's multitracks view and then match all the audio elements then go one by one and fix everything up.

[Chris Barnard]

You have to go to each source and normalize it.

Sure, I understand that. At the moment I am just making the quieter audio much louder and leaving the louder stuff (music etc) as it is. It seems to come out reasonbably well (Well, at least a lot better than before!) Any spikes in the quieter audio I am fixing microscopically.

It's good to know the nitty gritty behind it.

I guess what you are saying is that becuase of the inherent nature of audio, it is best handled manually than letting a computer normalise it, right?

[STORMDAVE]

I guess what you are saying is that becuase of the inherent nature of audio, it is best handled manually than letting a computer normalise it, right?

Exactly. Most NLE's can't handle audio properly, and you have much better tools in apps such as Audition2, or whatever else you can use.

If you're happy with how your audio is in Premiere 6.5 and by using it's tools, I don't see a reason why you would waste (I know this isn't the proper word) your time on doing everything externally, unless this project is really really important to you. Weddings and stuff, you shouldn't even really worry about, what you're doing is perfectly fine.

I usually use alot of filters in Audition when sweetening audio, especially if it includes speeches...I use compressors, EQ's, noise reducers all the time for this stuff. Also volume controls are much more precise in audio apps, such as sliders etc.

It all depends on the project, I do audio externally when the project is very important and if it's going out to the mass public. For events, I just try to stay at -2db and below, and that's about it.

[GrassValley_BH]

Yup. You want to make the "perceived" volume fairly constant. Otherwise you end up with the "TV commercial" effect where you adjust the listening volume just right then BLAM!! some commercial comes on blaring.

Music is a tricky one too. Just about all commercially-produced CDs have gone through some kind of "audio mastering" process which often includes compression/limiting, but even then you'll find that different CDs and even different tracks will "sound" louder or softer relatively.

A lot of it just ends up being experimentation until things "sound right" - even if the cold computer measurements are showing increases and decreases all over the place.

RMS measurements get closer to "how it sounds" but even then, you just can't beat listening to it yourself.

[STORMDAVE]

Good audio speakers are also needed to do audio sweetening...I have the KRK 5 Rokit speakers (As recommended by others) and the difference between these pairs and lets say a Logitech ones are night and day....it's like having a Sony PVM monitor for color grading and the same goes for audio.

Most editors ignore audio, but in my opinion audio can make or break a picture and it's just as important as video.

[GrassValley_BH]

Good audio speakers are also needed to do audio sweetening...

Agreed! Though I have to admit I had these 80-watt REVEAL (I've seen them badged with other brands) with a 3-pound power brick that really kicked ****...

I never took those puppies beyond 1/4 volume. Crazy. But now I'm a bit off-topic, so I will stop.

[Chris Barnard]

Sure, these are plain old weddings that I am doing. Nothing for public consumption. I'm not going to spend a massive amount of time on the audio. Just the discrepancies that were cropping up with the Ac3 conversion made the whole film sound (and look!) like a monkey had made it. I've noticed what Brandon said about Cd's as well, and usually I tinker with the volumes in most projects to get them to sound about right, and it seems to work quite well. I've good sennheiser mics so I never really get a problem with the initial quality of the live audio. But I think you might be right about the speakers. I've used ones other than my main ones and it was impossible to really get a feel for the audio. My main ones are not professional although they seem to do a decent job. Might have to spring for something like the ones you suggest.

Guys it's been an education - as always. Thanks!

[GrassValley_BH]

EDIUS's Audio Mixer "Touch" mode is your friend.

[biopic]

Actually there are those who would argue that you need to mix down using quality speakers, but then check the mix again using "domestic" ones (since in reality these are probably closer to what the listener will actually be using. In any case there is no right answer. I am told that in the UK the BBC gets more complaints about sound balances being too loud or too quiet, than any other single issue. Incidentally, can't wait until we get AAF export with Edius 4.5 - ideal for audio sweetening.

[GrassValley_BH]

Good point Andrew - it's always best to check on both what the "best" (or what the scopes) will see, and on what the majority of your viewers will see/hear.