16bit vs. 24bit Dialogue Recording
 

Is there any advantage to recording my dialogue in 24 bit, as opposed to 16bit? This would be for a horror movie.

+ safer level settings as you can leave more headroom
+ less rounding errors in complex mixdowns (mostly theoretical)

- takes up more space (not a big concern anymore)
- have to convert to 16 bits for editing anyway

Noise in a signal is additive when mixing several tracks together. If you record in 16bit, by the time you get to the final output you may find that you have a S/N ratio equivalent to having recorded in 14 or 12 bit. By recording and mixing in 24 and going down to 16 when you master, you avoid that issue. That being said, other factors in your audio chain might make such considerations moot anyway.

Do any of the current NLEs support 24bit audio tracks?

Oh yes, absolutely. Final Cut Pro, which is the one I'm familiar with, goes up to 96 kHz and 24 bits (more if you use Soundtrack Pro). I'd be surprised if this wasn't widely supported on Windows-based NLEs, too.

- Martin

Well, noise is additive of course when you SUM the signals, but then also the usefull signal is additive. For this reason the level of summed signals has to be lowered to prevent overloading*. Lowering the general level brings down also the noise. Noise and "pay" signals are no different, system does not differentiate them (in analog systems the noise adds up each generation, but it is history). So I venture to say noise does not add up. '

Using 24 bits does give some advantages in effects processing etc. because there are rounding errors in manipulated signals which might take away the last one or two least signifigant bits. In reality very few of us can make clean 24 bit originals anyway (Deva or SD 700 series or Cantar type field recorders and fine mics), so this is mostly theory for us like Steve says.

*) example: two tracks mixed 1:1 (summed) with peaks at -2 DbFS and noise at -70 DbFS would result in +4 dBFS peak -64 dBFS noise file, not possible. Have to lower levels 6 dB to keep peaks at -2 dBFS, this also pushes noise back to -70 dBFS.

Sony Vegas has supported 24/192 since the first version (nearly 8 years), even their cheapo Movie Studio supports it.
Adobe Soundbooth supports it.
I believe Canopus Edius supports it.

In other words, most NLE's support at least 24/96. But you're fine recording at 24/48, which I believe is supported by everyone.

You are thinking analog mixing here. Digital mixing, such as mixing music, dialog, and FX tracks in an NLE, is a different situation. There the noise level adds while the signal level does not, leading to an effective decrease in dynamic range to below that normally associated with 16-bit resolution. For best results, record in 24 bit even if final release is to be 16 bit.

Why does that happen?

I routinely record live music for distribution on either DVD or CD. I've found that 16 bit recordings don't have the dynamic range necessary for live recordings( I get a lot of clipping of vocals), in close proximity and in an enclosed space. 24 bit recordings really allow me to record live, without as much input compression or limiting, so, it saves me quite a bit of trouble during setup, not to mention adding compression in post to fix clipped passages.

It has to do with the math involved in digital signal processing amd I'm not enough of a mathematician to properly answer that - I only know that's how it works. Perhaps one our resident signal processing engineers will chime in and correct my crude explantion.

I think it goes roughly something like this ... When you mix digitally you are not summing signal voltages, you are adding numbers that represent the quantized waveforms. First, in binary arithematic 0+0=0, 0+1=1, 1+1=10, 10+1=11, 11+1=100, 11+11=110, etc. Now take two signals just at clipping, the top of the dynamic range spectrum. They are each represented by '111111111111111' (16 ones). Adding them should result in '10000000000000000' (1 followed by 16 zeros) but because 16 bits is the maximum, the combined signal is also at '1111111111111111', still 16 ones. But smaller numbers, down at the noise level, add just fine without that overflow and truncation.

If you are recording in 16-bit and are releasing in 16-bit without mixing multiple tracks, no problem. But mixing multiple 16-bit tracks costs you resolution for each added track. Of course there's nothing magic about 16 bit, the same loss happens when recording and mixing at 24 bits. But there, because you're going to reduce the resolution to 16 bits for release anyway, the loss of dynamic range that drops that of 24-bit originals down to that equivalent to 22 or 20 or 18 bit recordings is obscured by the final conversion drops it farther down to that of 16 bit.

Using binary does really make no difference, math is math.

I see it this way: we have two strong signals and sum them (mix them together). If there is a danger of going over the limit (be it 16 or 24 bits) we have to lower the levels of both signals before summing, because after summing and causing clipping they can not be saved. By deviding both signals by 2 (shifting one bit to the right) we make sure the result can not be larger than one original signal alone. This actually means lowering the levels 6 dB. This also lowers the noise levels one bit to the right or 6 dB. If we call noise the low level part of the signal that should not be there (hiss from amps, mic etc) it is also first lowered in level and in summing returned back to what it was. I can not see how this math would affect only loud part of the signal and not the quiet parts in equal effect. Thus I do not understand why only the noise would add up and anything else not. The system does not know which part of the signal is noise, which is not.

If we call noise the dithering etc stuff which lives down there among the last bit realm (lowest few decibels) my expertize can not figure what happens there, if that part behaves in linear fashion or do quantum physics come into play.

16 bit signal can carry 96 dB dynamic range, it is certainly possible to fit all music, not to mention dialogue in that space (and nobody has reproduction systems where the dynamic range including the listening space backround noise exeeds this). If the loudest signals clip the levels are set wrong. This does not mean using 24 bits for recording is vanity, it does give some peace of mind as you can set the levels lower for safety. As long as the mic preamps and mics are good enough, that is, many times thay are not. If there is too much hiss down there to prevent using 16 bits for recording without constant fear of clipping, using 24 bits is not going to help with same mics, pre-amps etc. The noise floor is going to be exectly the same distance from the clipping point.

Steve,

if you add '111111111111111' (16 ones in binary notation, or 65535 decimal) to itself (or multiply it with two, which is the same thing), the result is '11111111111111110' (16 ones and one zero, or 131070 decimal), which is way above the limit for 16 bits. The result you mentioned is what you'd get by adding just '1' to the original number, which is so darn close that - while technically clipping - it would sound alright.

If you sum up two tracks - digitally or analog - the result is naturally louder than the original tracks. If listen to two instruments, it is of course louder than just one of the two at a time. So if the tracks that you want to add up are close to the maximum level, you need to lower them before you add them, as Petri explained.

- Martin

If you have the option to go 24 bit, do it. With 16 bit make sure you have a good limiter and go for a hotter signal (-12 or so).
There is an interesting comparision of 16 and 24 bit on the Sound Devices Website (www.sounddevices.com). Listen to it. Bottomline (robustness in post left aside): for reasonably hot signals you will not notice much of a difference, but for lower levels its a huge step up with 24 bit.

In a practical sense, it won't make any difference. I'm guessing that your movie does not have a Hollywood budget and is a do-it-yourself kind of thing. Your biggest problem will likely be making sure there isn't too much background noise. Most locations you shoot in will have some sort of background noise to contend with. Usually the loudest noise will hide the quieter forms of noise.

*Background noise- for narrative work, this can be very high. This is by far your biggest problem.
*Preamp noise (sometimes this is louder than the quantization noise in 16-bit equipment, so there would be no point in going 24-bit; but some equipment is really good, so there may be a reason to go 24-bit)
*Noise in the playback equipment
*Quantization noise. Shooting 24-bit lowers this to negligible amounts.

If you have any ambiance in your mix, it will effectively mask any quantization noise. So I wouldn't worry about it at all.

The biggest problem with sound in independent films are (in roughly the following order):
*Bad script. (Ok this doesn't count.)
*Too much background noise
*Too much reverb

*Poorly-done ADR (and some actors aren't very good at ADR).
*Bad mixing; Lav-recorded sound doesn't match other dialogue
*Music not that good (Again, doesn't really count.)

2- To deal with background noise, the best way to record it would be to:
--Record boom audio. Indoors, use a hypercardioid microphone (not a shotgun, especially not the cheaper ones). A decent boom operator helps.
--On seperate tracks, record wireless lavs. Getting the mic so close will dramatically reduce background noise. However, you will need to EQ the lavs + add reverb to make them sound right. The lavs also cost money and suck up batteries.
Recording multi-track is easier with two audio people (one doing boom, the other mixing).

3- In music going 24-bit can help a bit... but that's when your background noise can get real low (relative to the signal, which is louder than dialog).