Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: 16 bit recording (Read 9119 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

16 bit recording

I often read how it's better to record music at 24 bits instead of 16 bits, so when processing the data the noise floor doesn't become audible (that's how I remember reading it). My question is one of quantification, in real world terms. Is there an estimate or rule of thumb on just how many transforms can be done on 16 bit recordings before the noise is generally audible? I figure there's too many variables to give any real answer, but I guess a typical figure would be worth something. Or perhaps something like, x transforms raises the noise floor y db, or something like that?

16 bit recording

Reply #1
I often read how it's better to record music at 24 bits instead of 16 bits, so when processing the data the noise floor doesn't become audible (that's how I remember reading it). My question is one of quantification, in real world terms. Is there an estimate or rule of thumb on just how many transforms can be done on 16 bit recordings before the noise is generally audible? I figure there's too many variables to give any real answer, but I guess a typical figure would be worth something. Or perhaps something like, x transforms raises the noise floor y db, or something like that?

What do you mean by transforms?

I think that the biggest motivation for using "as high as possible recording resolution" is the need for non-linear operations such as compressor/limiter and digital distortion effects. There is no theoretical limit to the degree of such operations that could be desired by the given sound engineer - but at some point the result will suffer due to limited source resolution

-k

16 bit recording

Reply #2
If one restricts the problem to transforms including proper non noise shaped dither, one can do 18 transforms+dither before the total noise exceeds -80dBFS. Every time you do a transform you add noise.

At least, if my calculation is correct. I assume a noise variance of (7/3)*2^(-2*NumberOfBits) for a TPDF dithering scheme.

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

16 bit recording

Reply #3
Doing transforms results in quantization errors. Their audibility, as distortion and/or noise, depends a great deal on the recording.

In some simple experiments I did some time ago, the quantization distortion for one transform became audible when the recording was at  -74dB and below. This is pretty far down if some of your signal is near 0dB, but if the ambient noise is very low, you may well be able to notice it. On the other hand, I’ve done possibly as many as ten transforms on 16 bit recordings from cassette without being able to tell any detriment, due to the tape hiss (even when low enough not to be objectionable vis a vis the recording -- i.e. the recordings were self dithering).

Dithering the transforms will eliminate the distortion, adding more noise in its place. I only have experience working with CoolEdit, which does not noise shape the dither under this circumstance. It only takes a few transforms on a good low noise recording before the dither itself starts becoming audible (Again based on my experiments. This will depend significantly on the recording). I found that if I stopped adding dither after about three transforms, the noise was not too too high and there was enough noise that no additional dither was needed.

16 bit recording

Reply #4
[...]I found that if I stopped adding dither after about three transforms, the noise was not too too high and there was enough noise that no additional dither was needed.

So, you can hear noise at -87dBFS? How loud do you listen to your music then?

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

16 bit recording

Reply #5
So, you can hear noise at -87dBFS? How loud do you listen to your music then?

Regards,
Jacco


in the mid-range - between 2 and 5 kHz psychoacoustics studies show that an SNR of 90 dB is not sufficiant to all listeners.
(i quote from the book - introduction to digital audio coding and standards - by M. Bosi and R.E Goldberg)

16 bit recording

Reply #6


So, you can hear noise at -87dBFS? How loud do you listen to your music then?

Regards,
Jacco


in the mid-range - between 2 and 5 kHz psychoacoustics studies show that an SNR of 90 dB is not sufficiant to all listeners.
(i quote from the book - introduction to digital audio coding and standards - by M. Bosi and R.E Goldberg)

That depends on the sensitivity of the listener (although it is quite hard to believe without assuming that the listening level must be > 85dBC SPL). But I am quite convinced that in music reproduction this large number of SNR is not relevant. If the noise floor is uncorrelated then it is more in the order of 70 - 80 dB (background noise).

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

16 bit recording

Reply #7
If I found a signal-to-noise ratio of 70 to 80dB adequate personally then I'd still be using my Teac V-1050 cassette deck with DolbyC. You don't have to be playing music at an ear-rupturing volume for a signal-to-noise ratio of that order to become an audible annoyance, particularly when listening with high quality headphones.

Cheers, Slipstreem. 

16 bit recording

Reply #8
how many transforms can be done on 16 bit recordings before the noise is generally audible?


Depends on the transform, really. If your transform is amplitude -70dB followed by amplitude +70dB, then the answer is 0 for 16 bit, but >0 for higher resolutions.
Ceterum censeo, there should be an "%is_stop_after_current%".

16 bit recording

Reply #9
It could be trivial to most, but you can convert 16 recordings to 24 bit (or 32, even floating point), do as many transforms as you need, convert the result back to the desired bitdepth with proper dithering.
If I'm correct, this would add as much noise as one transform done directly on the 16bit.
In theory, there is no difference between theory and practice. In practice there is.

16 bit recording

Reply #10
If I found a signal-to-noise ratio of 70 to 80dB adequate personally then I'd still be using my Teac V-1050 cassette deck with DolbyC. You don't have to be playing music at an ear-rupturing volume for a signal-to-noise ratio of that order to become an audible annoyance, particularly when listening with high quality headphones.

Cheers, Slipstreem. 

Well, I just did a check. I took white noise with an RMS level at -70dBFS and -80dBFS. I played some music (Within Temptation, at a level where my girlfriend normally shouts "Please turn it down!".) and as expected I could hear no background hiss. Without the music, I could hear a small portion of the -70dBFS but no -80dBFS. After cranking the volume up to levels close to insanity, I could hear something of the -80dBFS.

I did not try the -90dBFS. That would be rediculous (and dangerous to test with music).

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

16 bit recording

Reply #11
As I said, "...particularly when listening with high quality headphones.". You must have been playing it incredibly loud in your headphones for it to be at a level where your girlfriend would normally shout at you to turn it down!

Cheers, Slipstreem. 

16 bit recording

Reply #12
As I said, "...particularly when listening with high quality headphones.". You must have been playing it incredibly loud in your headphones for it to be at a level where your girlfriend would normally shout at you to turn it down!

Are you aware that blindtesting is necessary to post statements like those?

-k

16 bit recording

Reply #13
Apologies. I'll just go and poke my eyes out with a red-hot poker and try it again.

Seriously though, it's not too much of a problem when there's plenty of available signal to mask the noise obviously, but I've just spent the past two days trying to transfer a couple of audio books from Metal cassette recorded with DolbyC with a SNR of around 74dB and it's bugging the hell out of me during quiet passages. I guess that the same would apply to classical music with quiet passages too to an extent.

As a side note, I've reduced the problem in my particular instance by dragging an old homemade noise reduction system out of the cupboard and placing it between the cassette deck and the line-in on my sound card to increase the SNR by a perceived 10dB to around 84dB. It's a custom design based around an Analog Devices SSM2000 'Hush' noise reduction IC (datasheet) and can manage this level of improvement transparently, to my ears at least.

The reason for the minor disagreement we're having is most likely down to the fact that we're carrying out tests using totally different source material, differently 'shaped' noise sources, different listening equipment and different ears. We're obviously going to get very different results and I'm not disputing that noise at this level isn't a problem at all in your specific case.

I'm happy to accept that we're both right.

Cheers, Slipstreem. 

16 bit recording

Reply #14
Apologies. I'll just go and poke my eyes out with a red-hot poker and try it again.

Quote
8. All members that put forth a statement concerning subjective sound quality, must -- to the best of their ability -- provide objective support for their claims. Acceptable means of support are double blind listening tests (ABX or ABC/HR) demonstrating that the member can discern a difference perceptually, together with a test sample to allow others to reproduce their findings. Graphs, non-blind listening tests, waveform difference comparisons, and so on, are not acceptable means of providing support.


-k

16 bit recording

Reply #15
As I said, "...particularly when listening with high quality headphones.". You must have been playing it incredibly loud in your headphones for it to be at a level where your girlfriend would normally shout at you to turn it down!

Cheers, Slipstreem. 

That is a good joke! 

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

16 bit recording

Reply #16
@knutinh: I think you may be taking the TOS a little too literally. Neither myself nor anybody else can give evidence of a subjective situation until somebody comes up with a method of surgically grafting a person's auditory system onto somebody else's head and teleporting a sound system between locations so that everybody can hear it in exactly the same way.

It's not necessary for me to provide samples to other people for them to determine whether or not I can hear a noise component in a signal. I can hear it in this specific instance on my equipment with my ears. Improving the SNR so that it approaches 85dB makes the problem insignificant to me personally. It may or may not for somebody else.

We're all stating personal opinion in this thread so far. That's all it can be in the case of an ambiguous question like this.

Cheers, Slipstreem. 

16 bit recording

Reply #17
Thanks everyone for the replies. I figured it wouldn't be so simple but just wondered anyway as a practical matter. If I begin recording music of my own at home (rock style) with microphones (SM57s, & cheap Chinese Neuman imitation), what would the practical impact be of recording at 16 bit. Exactly how limited would I be in terms of applying effects, etc. in software before the final mix.

It could be trivial to most, but you can convert 16 recordings to 24 bit (or 32, even floating point), do as many transforms as you need, convert the result back to the desired bitdepth with proper dithering.
If I'm correct, this would add as much noise as one transform done directly on the 16bit.


If this is the case, then I guess I have nothing to worry about. Thanks again.

16 bit recording

Reply #18
Quote
I often read how it's better to record music at 24 bits instead of 16 bits, so when processing the data the noise floor doesn't become audible (that's how I remember reading it).

... If I begin recording music of my own at home (rock style) with microphones (SM57s, & cheap Chinese Neuman imitation), what would the practical impact be of recording at 16 bit. Exactly how limited would I be in terms of applying effects, etc. in software before the final mix.

...If this is the case, then I guess I have nothing to worry about. Thanks again.
  Remember there are two "ends" to your dynamic range.  I'd say the main advantage of 24 bits is that you can get more headroom without sacrificing resolution.  Headroom is important when recording "real-live" instruments and vocals, because the levels are not that predictable. 

From what I understand, pro software is often calibrated so that 0dB on the meters is actually -18dB Full-Scale (18dB of headroom).  That means you almost never have to worry about a loud vocal, or an extra-loud cymbal crash driving the level into clipping.    With 16-bits, you don't want to give-up 18dB of range/resolution, so you sometimes end-up pushing levels to the point where you get clipping. 

This extra headroom also comes in handy when you mix (since mixing increases the level).  Although, you can get-around that problem easily by mixing from 16-bit to 24 bit.

16 bit recording

Reply #19
Is it really worth debating whether 16-bit would be sufficient, when 20- and 24-bit converters are cheaply available?

16 bit recording

Reply #20
This really isn't so much a question of the resolution of recording than the resolution of the processing which the OP did not specify.  I'm hard-pressed to think of a situation these days where one is forced to process a 16-bit sample at 16 bits.  You can process a 16-bit sample all day long at 24 bits with minimal degradation in terms of SNR so long as you only go back to 16 bits when you're finished, can't you?

16 bit recording

Reply #21
You can process a 16-bit sample all day long at 24 bits with minimal degradation in terms of SNR so long as you only go back to 16 bits when you're finished, can't you?


Even better, you can process at 32 bit float and be safeguarded also against clipping from exceeding 0 dBfs on intermediate stages.
Ceterum censeo, there should be an "%is_stop_after_current%".

16 bit recording

Reply #22
You can process a 16-bit sample all day long at 24 bits with minimal degradation in terms of SNR so long as you only go back to 16 bits when you're finished, can't you?


Even better, you can process at 32 bit float and be safeguarded also against clipping from exceeding 0 dBfs on intermediate stages.

As long as you are performing quasi-linear operations, then yes.

If you are doing heavy non-linear processing, then no matter the intermediate resolution, the resolution of the recording could be an issue.

Imagine amplifying parts of the signal by +60dB. Sound engineers could be using compressor, limiter, distortion or any other effect to achieve their artistic goal.

They could also be recording at conservative levels to avoid clipping if the source was very dynamic (or not used to singing into microphones).

-k

 
SimplePortal 1.0.0 RC1 © 2008-2021