What to trust, the meter or my ears?
 

During mixing, what should I do if the meters indicate that an event should be loud (say, peaking +24dB above dialnorm) but it sounds fine on my mediocre setup? Does that mean it will sound horrendously loud in a theatre? Should I look at the RMS value instead? I am talking about really short events, like closing a cell phone or a shelf.

The short answer: I would use the meters as a guide when it comes to peaks. Just becuase it sounds good on your set up with peaks just below 0 db does not mean that something down the line is not going to distort when it comes to the peaks.

The longer answer: While some non-linear video editing systems use -12 dB as the reference standard for peaks, you'll find that most professional audio systems use -20 dB as the standard reference. We're taking digital here were 0 dB is the brick wall and beyond that, raspy, horrble noise will result;. It's often debated if peaking at -12 dB is too hot and -20 dB should be used as the reference. The issue is that if you go over -12dB peak, you might have a problem with distortion if you go to an analog format like Betacam SP (assuming you line up -12db on your NLE to 0db on the analog system). While 0db in digital is a brick wall, 0db in analog still allows for some headroom with +3db peaks not considered a problem. As long as you have good dynamic range, don't be concerned about much of the program being below your peak levels, but you should be concerned about too much peaking.

If you notice your video does not sound as loud as a commercial video, yet your peak meter readings look right, don't give in to the temptation to make the video hot with too many peaks. Instead, you may need to apply a little bit of compression to the audio so the quiet passages are louder while the peaks are a little less peaky. How to apply compression is described in excellent detail in the book "Audio Postproduction for Digital Video" by Jay Rose.

Thanks for the advice! I have that book, and I did apply some compression, but I was surprised that a +20dB excursion did not sound as loud as I thought it would, so I thought maybe the length of the event is an important factor.

If the signal to be recorded meets certain criteria found only in textbooks the proper rms load for an A/D converter is about (depends on the number of bits) 13 dB, or if its a pure tone such as a solo French horn, 3dB, below the rail (0dBFS). At that load (under those conditions) the overload light on each 48kHz sampled channel would blink on average once every 2.5 seconds or so. What a recordist in the real world wants to do is to set gain to the point where the overload light blinks infrequently and then back off say 3 dB from that point so that it never blinks again. With 24 bit recording you will cover the dynamic range of most mics. With 16 bit you may have to manage gain as the level of the incoming material changes.

The meters themselves determine how to read them depending on their 'ballistics'. The most informative type goes instantly to the highest level it sees and hangs there for some period of time or falls back slowly so that you catch the fact that a peak was recorded. Others (especially in software) will light an overload light and keep it lit until you reset it or for some predetermined period of time. Other types mimic the old mechanical meters and follow the average or rms (not the same) signal level. The Tascam HD-P2 has the neatest display I think I've ever seen. The overload "light" (it's really an LCD display segment) is there and stays "illuminated" for a couple of seconds after a peak. A "ball" rides the peaks and falls gradually back to a bar which follows the rms input. With this kind of display you can fairly easily find a setting which prevents you from hitting the rail and that's the thing to do (IMO).

What NLE are you editing with? If you're using FCP, for example, you can have it "find the peaks" so you can decide what to do wth them. If you just have a few peaks here and there that go over, you also don't want to bring the whole program down too far just because of some peaks. In some cases, the thing to do is to deal with the specific peaks, so the average programs levels are "meaty."

I edit with Vegas, and I have SoundForge too. The problem is not clipping--the recording is fine; it is whether something that looks wrong but sounds right on my setup will sound wrong on another system, esp. a theatrical one.

[QUOTE=David Tames]
The longer answer: While some non-linear video editing systems use -12 dB as the reference standard for peaks, you'll find that most professional audio systems use -20 dB as the standard reference.

***STOP! the -20 dB you refer to as standard reference is for a 1 kHz tone, not peak audio. You set tone at -20 and have room from -20 to 0 for your peaks.

It's often debated if peaking at -12 dB is too hot and -20 dB should be used as the reference. The issue is that if you go over -12dB peak, you might have a problem with distortion if you go to an analog format like Betacam SP (assuming you line up -12db on your NLE to 0db on the analog system). While 0db in digital is a brick wall, 0db in analog still allows for some headroom with +3db peaks not considered a problem. As long as you have good dynamic range, don't be concerned about much of the program being below your peak levels, but you should be concerned about too much peaking.

***Again, -20 dB reference is NOT for peak, it's for solid tone.

Your transfer example is an entirely different situation. In the case of transfer from digital to Betacam SP, if you had set tone at -20 dB on the digital system and knew that you had peaks as high as 0 dB, you'd use that digital -20 dB tone to set the record level of the Beta SP to maybe -14 dB (assuming the Beta SP recorder could take + 6dB peaks.)

Regards,

Ty Ford

BTW,

sharp transients (peaks) are created when someone drops something, claps their hands, bangs a hammer, shuts a door, opens a three-ring binder.

It is not at all unusual to see spikes in the waveform when these event happen. You should record with enough headroom so they don't clip, but because they are usually VERY brief, the clip is hard if not impossible to hear.

A mixer with a good limiter will prevent most "overs" like this, but even they may not be fast enough to catch them all.

Regards,

Ty Ford

Ty must have seen my film, because the events I am referring to are the very ones he mentions. Am I to understand that sharp transients are alright, as long as they do not clip; i.e., I should trust my ears?

Yes, I think so. I was pretty amazed the first time I ran into transient like the ones I mentioned. Wow! what peaks! No you understand why the director and actor work out all of the moves so the actor doesn't drop a book on his or someone else's lines. Listen (and watch) for that next time to watch a movie.

Regards,

Ty Ford

I am working on a project now with lots of peaks like you mention in the dialog. Some are from ess sounds, some are from the actors clicking there tongue (tsk) before a line. I am matching hundreds of short takes, and it is difficult to judge by the meters. I use Pro-Tools, and my first step has been to manualy highlight these peaks (or cutting sometimes) and then apply a gain reduction, then normalize until all the bits are about the same level. Tedious, but it works well.

Overly sibilant actors should go to a speech therapist and get that taken care of. I'm an AFTRA/SAG acotr, as it turns out. I teach narration and voiceover. When someone comes to me with excessive sibilance, I direct them to a speech therapist before we begin working together.

Some of them come from stage and say, "(tsk) Well nobody elssss ever ssssadid I wasss too ssszzzibilant!" I tell them that on stage it doesn't matter as much, but when you're doing work into a mic, it mattters a lot.

Most have taken my advice and have thanked me for helping them. A few have resisted and still suffer with the problem.

BTW, if teh tsks are not in the script, or part off teh role, direct them not to do it.

Regards,

Ty (brutal director and voice coach) Ford

Ty, thanks for pointing this out, I left an important paragraph out of my discussion, Yes, the -20 or -12 is where you place tone or where your "average" signal hovers around, I did not mean to imply that's where the peaks go, but rather where you place your "average" signal and how much dynamic range or headroom you're allowing yourself. I've had problems down the line when I had a good portion of the signal right up to 0 most of the time, so I try to avoid having peaks over -3 as a safety zone to play it safe. But yes, of course, peaks should go right up to (but not over) 0. The -20 vs -12 thing is also related to the issue around how much "headroom" you want to allow for your project. I'm not wording this very well, thanks for your clarification, a very critical one!

No Problem. The devil is in the details.

-3dB is usually where I stop, depending on where the material is headed.

Some (most) satellite uplinks, so I'm told, don't want any peaks louder than -12 dB. Apparently these particular satellites don't have much headroom.

Network/cable TV is a little higher, but not much. If you are preparing a show for Discovery or another network, it's ALWAYS advisible to ask for their specs.

Regards,

Ty Ford

Ty, let me rewrite that so it actually makes sense (what was I thinking?) and does not mislead anyone, please tell me if the following paragraph makes sense now (consider the following a draft for comment):

It's often debated if your average signal should hover around -12 db or -20 db, it's a question of how much headroom you want to allow for the peak portion of your program, which should not go over 0 db. It also has lots to do with the overall dynamic range of your mix. As long as you have a good signal with peaks and valleys that average around -12 dB with plenty of peaking up the -3 db range and a few right to 0 db (it's good to use a limiter on the final mix), don't be too concerned about some of the program being closer to -12 db than 0 db, as long as the average hover just over -12db. -20 average makes sense for production recording, but for a final mix you're better off with -12db average and a healthy number of peaks in the -6db to -3db range.

A separate issue is dealing with other formats downstream, one issue is that if you have most of the peaks right up against the 0 dB brick wall all of the time, you might have a problem with distortion if you then go to an analog format downsteam like Betacam SP (assuming you line up -12db reference TONE on your NLE to 0db analog on the analog system). 0db analog still allows for quite a bit of headroom with +3db analog peaks not considered a problem, in fact it is good to see a reasonable number of peaks over 0 dB analog.

Ty, please, set me straight here.

Hi David,

Pretty much. I use a combo meter that shows RMS and peak simultaneously. The meter has a "peak hold" feature that indicates where that peak hit for a few seconds before going back down. That's helpful in case I'm looking away for a moment.

I generally don't pay much attention to the RMS level after I set levels with a reference tone. I DO pay a lot of attention to the peak level.

Regards,

Ty Ford

Thanks Ty, DVi is a really awesome board, where else do things like this get straightened out so quickly, efficiently, and accurately, like this?