16bit vs. 24bit Dialogue Recording

[Steve House]

Originally Posted by Martin Pauly

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

Thanks - I knew the number didn't look right as I typed it but I was too sleepy to figure out the right one. Nonetheless, I'll still stick to my guns ... digital mixing can lead to a loss of effective dynamic range and recording at a higher bit depth can compensate for that. See also this discussion on digital recording in "Sound for Film and Television", Thomlinson Holman, page 64, where he writes "With any degree of mixing (of multiple 16 bit tracks) it is impossible to deliver an output that has a 16 bit dynamic range."

[Bob Grant]

Just one small correction, the maximum value in a digital recording is represented by 0, not (2^n) - 1 where n is the bit depth. This makes for much simpler conversions between bit depths.
Certainly mix at 24 bit will be better than mixing at 16 bit however it's very common practice for modern mixinf apps to run their internal pipelines at more than 16 bits anyway so unless you're render back to 16 bit through multiple generations you don't gain anything.

[Ty Ford]

Of course the dialog world got along quite nicely with 16-bit, 48 kHz DAT machines before becoming obsolete.

Regards,

Ty Ford

[Petri Kaipiainen]

And before 16 bit DAT hissy analog tape was good enough...

Getting the most advanced gear and technology does not help if techique is not right.

And in the end many videos/films are delivered on DVD:s with MPEG sound not even 16 bit quality...

[Ty Ford]

Finland, Australia, Canada, USA, etc. What a great forum!

16-bit for dialog is fine IF..notice the big if...the audio is particularly well recorded. However, for some members and visitors, audio is a necessary, mysterious PITA.

They can't afford the gear or a sound person and/or do not understand "best practices" to achieve well recorded audio themselves. For them, 24-bit may afford them some safety margin if there's someone who knows how to dig out the dialog with intelligent use of good EQ. In most cases, though, they aren't going to spend the time or money, so the point is moot.

I'm not trying to sound harsh here, but although I'm driven to get my clients the best possible audio (because that's what they pay me to do) many forum members and visitors aren't making a living with video and audio. For them, really good audio (or video) is not as important as it would be for me.

The MPEG audio used in HDV cameras is an interesting example. It has 1/5 the data of 16-bit 48 kHz audio. Some folk think that's fine!

Regards,

Ty Ford

[Steve House]

Originally Posted by Ty Ford

...
16-bit for dialog is fine IF..notice the big if...the audio is particularly well recorded. ...
I'm not trying to sound harsh here, but although I'm driven to get my clients the best possible audio (because that's what they pay me to do) many forum members and visitors aren't making a living with video and audio. For them, really good audio (or video) is not as important as it would be for me.

The MPEG audio used in HDV cameras is an interesting example. It has 1/5 the data of 16-bit 48 kHz audio. Some folk think that's fine!

Regards,

Ty Ford

Agree with you 100%, Ty.

[Glenn Gipson]

I have a lot of yelling and screaming in my movie, will 24 bits better for such loud signals?

[Steve House]

Originally Posted by Glenn Gipson

I have a lot of yelling and screaming in my movie, will 24 bits better for such loud signals?

The dynamic range we're talking about is the ratio between the softest sound possible and the peak levels. Screaming and yelling are better controled in the analog realm with attention to level setting and proper application of limiting. Once the audio is pristine, there *might* be a subtle difference between 16 bit and 24 bit but going to 24 won't be a panacea for out of control levels.

[Thomas Johansson]

Hey there, sorry to bring up an old thread but I'd like to add a few points to the 16 vs. 24 bits here. I've worked as a mastering, mix engineer and producer for the last eight years and just getting into video for the fun of it... so I'm a video newbie but an audio expert.

Something that a lot of guys don't realize, even in the audio industry, is what actually happens with your audio in the DAE. Always remember to apply the correct dither at the end of your audio chain. It doesn't matter if you have 24bit, 16bit or 8bit audio at the source. As soon as you start manipulating the audio in your DAE, using panning, volume, plugins, fades etc... you have 32bit float at your output bus. Just lowering a channel 0,1db means re-calcualting the audio, and to keep it as close to the original as possible the software needs 32bit float resolution or higher. If you have an audio project that you're mixing down to 16bit without adding dither noise at the output, it will sound horrible compared to the same project properly dithered. If you've ever had the experience of fading out a piece of audio, and that it didn't sound totally smooth... like at the end of the fade it sounds more like the sound just drops out rather than smoothly disappear, you've not dithered your audio correctly!

ALWAYS REMEMBER TO DITHER!!!!!

My take on dialogue recording is that it might of course sound better at 24bit compared to 16bit, but it also depends A LOT on the quality of your mic, preamp and A/D converters. 24bit also sounds a lot better if you need to apply heavy processing at some point, especially dynamic processing. There really is no excuse for NOT recording at 24bit if you can, just remember that unless you're using only this recording in your final production totally unprocessed... you will always have the need for DITHER!

[A. J. deLange]

Let's say we have a console with some three bit A/D converters with step size 1 volt (to make the math easy). If these converters are "offset binary" they will produce code 000 for -4 volts, 001 for -3; 010 for -2; 011 for -1; 100 for 1; 101 for 2 and 111 for +3 volts which is the largest positive number which can be represented with 3 bits and 1 volt per count. When these numbers are to be used for arithmetic operations they are converted to "2's complement" by changing the most significant (leftmost) bit so now -4 volts encodes as 100; -1 volt as 111; 0 volts as 000; +3 volts as 011 and so on. Note that these numbers are representations of the voltages at the input to the A/D which voltages are proportional to the air pressures at the microphone diaphragms. Thus the numbers are scaled approximations to sound pressure levels over time.

Now if we want to sum the output of 2 of these A/D converters we may be summing -1 with +3 to get +2 or +1 with +1 to get +2 or -2 with -2 to get -4 but there are equally good chances that we will be summing +2 with +3 to get +5 or -3 with -4 to get -7. Neither of these last 2 is representable with 3 bits so we must go to a 4th bit and indeed if we know we are summing three bit "signed integers" we will recode the numbers by duplicating the sign bit thus -4 becomes 1100; -3 = 1101, -1 = 1011; 1 = 0001 and so on. Now if we need to add -3 to -4 we write 1101 above 1100 and add the bits by the binary rules (1 + 0 =1; 1+1 = 0 carry 1) so 1101 + 1100 = 1001 (the carry into the 5th binary place gets discarded) which is -7 in 4 bit 2's complement notation (-6 is 1010 -1 is 1111; +7 is 0111...). So in adding 2 3 bit channels we need a 4th bit. If we were to add 2 of these 4 bit sums of 2 we would need a fifth; adding 2 sums of sums a fifth and so on. Thus every time you double the number of channels you need an extra bit but note that this is the most conservative approach. Signals and noises are usually uncorrelated meaning that it is the square of the voltage of the sum which doubles if two signals are added. Nevertheless the math requires the extra bit and it is therefore supplied. To require 24 bits as a consequence of summing 16 bit channels means that up to 256 channels could be accomodated.

If you sum 256 channels of 16 bits each forming a 24 bit sum and then go back to 16 bits by truncation or even rounding that is akin to quantizing the original signal to 16 bits and quantizing noise will be generated. It will be 10.8 dB below the least significant bit but if the signal is narrowband (unlikely from a sum of 256 sources) it will be correlated with the signal and you can hear it. Dithering decorrelates the quantizing noise from the signal but with a mix of 256 sources that shouldn't matter much. Dithering is also beneficial in this regard in the hardware A/D converters.

[Roshdi Alkadri]

hi, i also wanted to chime in here. I use to ask the same question about 16/24bit. After a few productions and recording sound seperately i realized the huge difference. 24 bit gives you less noise and higher headroom, you also have more freedom when processing the files with filters and effects. When i mix my productions, there's so much room to play with. watch this weird short film we did and notice the yell at the end, its so clean and undistorted, there was also no gain riding at all. All SFX were 24bit too. Recored using the 416 mic and the Sound Devices 702T.

http://www.youtube.com/watch?v=jFWP1hGORis

[A. J. deLange]

Originally Posted by Ty Ford

The MPEG audio used in HDV cameras is an interesting example. It has 1/5 the data of 16-bit 48 kHz audio. Some folk think that's fine!

Ty Ford

And much of the time it is. If the entropy (a mathematical measure of the information content in a signal) is less than 16/5 bits per sample it is theoretically possible to convey all the information in the signal with 3 1/5 bits (average obviously) per sample. In other words compression in and of itself is not necessarily a bad thing. Most sources can be compressed by about 50% losslessly meaning that a 16 bit signal reconstructed from a 8 bit compressed samples is identical to the 16 bit original). The problems occur when "compression artifacts" arise because of unusual patterms in the sound (and implementations of lagorithms that don't realize their theoretical potentials). It's sort of like the extra bits required for summing situation. Most of the time you don't need the extra bit but when you do, you do.

[Ty Ford]

Spoken like a mathematician. :)

Herb Squire's test at a past AES conference in NY audibly demonstrated the bad effects of of data compression on audio. I was there. I heard it. The eye is a lot more forgiving than the ear when it comes to data compression.

Further, depending on the use of the audio, subsequent compressions - like when you make a regular DVD and the audio is then compressed as part of the mpeg stream - more data is lost.

Were we able to hear both the original audio (recorded well) at 24-bit and then be able to compare that with HDV audio, that would be telling. I recall some folks double recording (to camera and to 702T) and letting us know the 24-bit sounded better. Of course, part of the problem could have been the camera audio circuits.

In a audio/video world where 24-bit audio is normal (and 32-bit is not unusual), the compromises made by the industry to make and sell HDV are questionable. But then VHS won over beta and there are still a lot of audio cassettes out there. If recorded well, those cassettes sound better than many of the XM broadcasts due to XM's data compression.

Let me decide when I want to chuck the bits, A. J.. :)

Regards,

Ty Ford

[A. J. deLange]

Ty,

I'm not a mathemetician but rather and engineer which in some quarters would be deemed even worse. That confession made let me assure you that I am the last guy to try to impose bit chucking on you because I am well aware of the problems that arise when the theoretical isn't well mapped into practice. It seems that in any compressed piece that I listen too (from I-tunes for example) there is always at least one nasty artifact which usually occurs on some bit that should be brilliant like a trumpet blast, symbol crash or high C even though the rest of the performance is fine (to my tin ear anyway). I, and some of my colleagues, have been wondering what the future of recorded music is to be now that CDs seem to be fading from the scene and SACD's looking as if it isn't going to make it.

Nevertheless, if I wanted to send you an (uncompressed) audio file I would certainly run it through Stuffit (or some similar compression routine) first in order to save time and bandwidth on the Net. When you received this file you would un-stuff it (expand it) and receive an identical, to the bit, copy of the original file. This is an example of lossless compression. The compressor doesn't care whether the data is video, audio, pressure data, text or computer source code. It analyses the 1's and 0's and codes the data in such a way that the number of 1's and 0's sent is about the same as the true information content of the file.

[Thomas Johansson]

A.J says it beautifully, and probably correct!

Thing is, the ear is very sensitive to artifacts. I can't understand how people can listen to 320kb MP3-audio for instance, and I'm still far away from being an audio snob. But when you combine video and audio you're also dividing your attention. For the most part, audio will seem to sound better when you're also focusing on what you see (a well known fact amongst top-plugin developers) so you can get away with a lot more than if you were focusing on audio only. Even music videos seem to enhance the production quality of music by a great margin. You might see a video and be impressed by the audio production, but when listening to the audio only you might decide that it was really quite un-impressive.

You can get away with more audio compression than usual if you're combining it with video. I typically don't react to 224kb mp3 audio sounding bad when also focusing on video, but somewhere around there is the limit for me.

16bit vs. 24bit recording then becomes kind of silly to discuss since if you're not bothered by the former, you will most certainly not be bothered by the latter. Most people don't even react to a truncated 16bit audio stream, and given the right situation and production I would probably not react to it either since it only really has a musical impact when the sound becomes more dynamic. Most non-audio people would react to a truncated classical recording once shown what to listen for I'm sure, but most people including professionals, would not react to it if it were rock, metal or smashed RnB which is essentially very non-dynamic and distorted by nature so to speak.

To be on the safe side... always record at the highest resolution and always dither for your end-medium. You never need to think about it... just do it! :)