camn
Active member
Suppose I converted a bunch of tracks from 16bit 48khz UP to 32-bit. Would it improve the sound, degrade the sound, or do nothing..?
xoxo
cameron
xoxo
cameron
dobro
Well-known member
You can't do it, man. It's one of those facts of life - you can't unboil an egg or get somebody unpregnant, and you can't convert bitlength up. Once you've recorded in 16-bit, that's it. You can convert *down*, sure, but that's another story.
E
Evgueni
New member
Dobro is right - in case that upconverting was going to be your final result.
If, on the other hand, you plan on doing lots of processing on these files, then upconverting makes sense. This way you won't accumulate truncation errors, and the final result will sound better than if you'd do all transformations in 16 bits. Even if you then downconvert the result to 16 bits.
If, on the other hand, you plan on doing lots of processing on these files, then upconverting makes sense. This way you won't accumulate truncation errors, and the final result will sound better than if you'd do all transformations in 16 bits. Even if you then downconvert the result to 16 bits.
dobro
Well-known member
You mean carry out all CEP operations in 32-bit mode, right?
camn
Active member
THAT"S what Im talking about.. converting up.. doing all my mixing and processing, then back down.
Better than doing it in 16? Because it CAN be done. physically, that is.
xoxo
cameron
Better than doing it in 16? Because it CAN be done. physically, that is.
xoxo
cameron
dobro
Well-known member
My understanding's this: you can't convert 16 bit to 32 bit, it's like trying to get a high resolution photographic print out of low resolution film. But I've heard that it's useful to carry out various editing operations in 32-bit mode, which you will of course have to convert back to 16 bit if you want to burn a CD.
E
Evgueni
New member
Yes, it's better to do all the processing in 32-bit mode, even if your original files are 16 bit ones.
Here's a simple example.
You know that in 16 bit words, you have the samples represented by integer numbers ranging from -32767 to +32767.
Say you have two consecutive samples - one of the value "4" and another of the value "6". Then you decide to amplify them by 10%, and you do it four times.
So, after the first transform, the real value of the first sample should become 4.4. But in 16 bits this cannot be represented. So it gets rounded to 4. Next transform gets this 4, should make it 4.4, but it again has to be rounded to 4. So, after 4 transforms,
we'll have the sample equal to 4, while in reality the actual number must be 5.8564 (I've used calculator for this), which would be rounded to 6.
Now, let's see what happens to the next sample, which in the beginning was equal to 6. After the first transform, 6 and 10 percent should become 6.6, but it has to be rounded to the nearest integer, so it becomes 7. Next transform makes it 8 (7.7 rounded). Third one - 9 (8.8 rounded). Fourth one - 10 (9.9 rounded). But the real value (if no roundings happened) would be 8.7846 (again, I used calculator to come up with this number).
So, before all transformations the second sample was 1.5 times larger than the first one (6/4). As we've seen, after all transformations the first sample will still be equal to 4, the second one - to 10. That is, the second sample becomes 2.5 times larger than the first one! That's gross distortion, and, obviously, not what you'd expect to get.
In 32 bit words, you may represent those non-integer numbers more exactly. So, after those same 4 amplifications, you'd have the first sample value of 5.8564 (see above), and the second one's - 8.7846. When you'd converted the file back to 16 bits, those would become 6 and 9 - much more exact representation of what had to be!
Here's a simple example.
You know that in 16 bit words, you have the samples represented by integer numbers ranging from -32767 to +32767.
Say you have two consecutive samples - one of the value "4" and another of the value "6". Then you decide to amplify them by 10%, and you do it four times.
So, after the first transform, the real value of the first sample should become 4.4. But in 16 bits this cannot be represented. So it gets rounded to 4. Next transform gets this 4, should make it 4.4, but it again has to be rounded to 4. So, after 4 transforms,
we'll have the sample equal to 4, while in reality the actual number must be 5.8564 (I've used calculator for this), which would be rounded to 6.
Now, let's see what happens to the next sample, which in the beginning was equal to 6. After the first transform, 6 and 10 percent should become 6.6, but it has to be rounded to the nearest integer, so it becomes 7. Next transform makes it 8 (7.7 rounded). Third one - 9 (8.8 rounded). Fourth one - 10 (9.9 rounded). But the real value (if no roundings happened) would be 8.7846 (again, I used calculator to come up with this number).
So, before all transformations the second sample was 1.5 times larger than the first one (6/4). As we've seen, after all transformations the first sample will still be equal to 4, the second one - to 10. That is, the second sample becomes 2.5 times larger than the first one! That's gross distortion, and, obviously, not what you'd expect to get.
In 32 bit words, you may represent those non-integer numbers more exactly. So, after those same 4 amplifications, you'd have the first sample value of 5.8564 (see above), and the second one's - 8.7846. When you'd converted the file back to 16 bits, those would become 6 and 9 - much more exact representation of what had to be!
dobro
Well-known member
Got that Camn? 
Hmm... impressive what you can do with a bit of knowledge and a calculator. What this says to me, aside from the obvious "do everything in 32-bit" is "keep the number of transforms to a minimum".
Do you track using compression, by the way?
Hmm... impressive what you can do with a bit of knowledge and a calculator. What this says to me, aside from the obvious "do everything in 32-bit" is "keep the number of transforms to a minimum".
Do you track using compression, by the way?
camn
Active member
Got it.
In a way, I track using compression. Id also love everyones thoughts on this.
I track dry, and for comression I run tracks through an outboard 24-bit digital compressor. The same for effects. (I do It so I can have realtime control over things using knobs and stuff.) I figure If I used the same proceesors to track with effects, the result would be Identical since Im staying digital the whole time.
thoughts?
xoxox
cameron
In a way, I track using compression. Id also love everyones thoughts on this.
I track dry, and for comression I run tracks through an outboard 24-bit digital compressor. The same for effects. (I do It so I can have realtime control over things using knobs and stuff.) I figure If I used the same proceesors to track with effects, the result would be Identical since Im staying digital the whole time.
thoughts?
xoxox
cameron
dobro
Well-known member
When I said 'track using compression' I meant this hookup: mic > pre > com > recorder. So if you're tracking dry, I guess you're not doing that. Reason I ask: tracking with compression alters the signal, as does tracking with any effect. I track dry, and then consider effects. But I've heard that engineers who really know what they're doing track with a certain amount of compression, as I've described above, and I'd like to know exactly what it accomplishes, aside from being able to record a hotter signal without clipping. I'd also like to know if there's any kind of noticeable degradation of the sound as a result.
Evgueni?
Evgueni?
E
Evgueni
New member
Depending on situation, there may or may not be some benefits in this.
If you know that you will add some compression to that particular track anyway, doing it the way you described might actually be a good idea because:
1) if your recorder is digital, then not having to compress later in digital domain reduces the rounding errors.
Those errors, obviously, affect softer sounds more that the loud ones. In my example above I deliberately selected very small values of samples (4 and 6) to make the resulting error gross compared to the sample values themselves; it's kind of worst-case scenario. If I selected the values of, say, around 10000, then the relative error would be much less noticeable.
So, if you have your track recorded hotter, then any subsequent digital transformations (including EQ, compression, normalisation, amplification etc.) will result in less noticeable distortions.
2) If your recorder is the analog one, then if you compress whatever is already recorded on tape, you'll get "noise pumping". That is, the tape noise in the softer passages will get louder than in the louder ones. This may or may not be noticeable depending on the tape noise level.
3) Many people beleive that their analog outboard compressors are of better quality than whatever digital effects, plug-ins etc. they have. Again, it may or may not be so in your situation.
==================
(BTW, I suspect my English makes whatever I try to explain to look more difficult than it really is; sorry for that. These things are really simple (at least in theory), it's only the explanation that is difficult).
If you know that you will add some compression to that particular track anyway, doing it the way you described might actually be a good idea because:
1) if your recorder is digital, then not having to compress later in digital domain reduces the rounding errors.
Those errors, obviously, affect softer sounds more that the loud ones. In my example above I deliberately selected very small values of samples (4 and 6) to make the resulting error gross compared to the sample values themselves; it's kind of worst-case scenario. If I selected the values of, say, around 10000, then the relative error would be much less noticeable.
So, if you have your track recorded hotter, then any subsequent digital transformations (including EQ, compression, normalisation, amplification etc.) will result in less noticeable distortions.
2) If your recorder is the analog one, then if you compress whatever is already recorded on tape, you'll get "noise pumping". That is, the tape noise in the softer passages will get louder than in the louder ones. This may or may not be noticeable depending on the tape noise level.
3) Many people beleive that their analog outboard compressors are of better quality than whatever digital effects, plug-ins etc. they have. Again, it may or may not be so in your situation.
==================
(BTW, I suspect my English makes whatever I try to explain to look more difficult than it really is; sorry for that. These things are really simple (at least in theory), it's only the explanation that is difficult).
camn
Active member
soo, I run my signal through a A/D converter and record exactly the data that that converter produces. Then I run my recorded signal through a digital compressor and rerecord. Is it equivalent to running my signal through my D/A, then through the same compressor, then recording?
just curious,
cameron
just curious,
cameron
E
Evgueni
New member
Camn: Yes, if you always transmit your signal in digital form after A/D, and if your compressor operates at the same word length and quantization frequency as your recorder and A/D, it should be equivalent.
But if, for example, you have A/DC and comressor working at 24bit word length, and the recorder working at 16bit (so you have to downconvert before recording), then the sequence A/D->recorder->compressor->recorder would obviously produce more distortion than the sequence A/D->compressor->recorder.
(Note that what dobro talked about was something different - he meant compression in the analog domain, before A/D)
But if, for example, you have A/DC and comressor working at 24bit word length, and the recorder working at 16bit (so you have to downconvert before recording), then the sequence A/D->recorder->compressor->recorder would obviously produce more distortion than the sequence A/D->compressor->recorder.
(Note that what dobro talked about was something different - he meant compression in the analog domain, before A/D)
dobro
Well-known member
Nope... sorry... when I said 'recorder' I meant anything that records - digital or analog.
I'm interested in how some engineers use compression as part of their tracking/recording chain - they compress the signal from the mic+preamp before it goes to hard disk or tape. Engineers are interesting people to pay attention to because they have a lot of experience of *listening* to music and the recorded final result - in other words, there's a good chance they know what they're doing when they compress the signal on the way to the recorder, rather than afterward, when it's being mixed. You say:
"not having to compress later in digital domain reduces the
rounding errors."
I'm grateful to you for showing me something I didn't know before (yes, I'm concerned with avoiding or reducing processing 'errors'), but what I'm really curious about is how it sounds, and why so many engineers use compression when tracking.
This has nothing to do with CEP, unless you can think of something.
I thought you were a native speaker.
I'm interested in how some engineers use compression as part of their tracking/recording chain - they compress the signal from the mic+preamp before it goes to hard disk or tape. Engineers are interesting people to pay attention to because they have a lot of experience of *listening* to music and the recorded final result - in other words, there's a good chance they know what they're doing when they compress the signal on the way to the recorder, rather than afterward, when it's being mixed. You say:
"not having to compress later in digital domain reduces the
rounding errors."
I'm grateful to you for showing me something I didn't know before (yes, I'm concerned with avoiding or reducing processing 'errors'), but what I'm really curious about is how it sounds, and why so many engineers use compression when tracking.
This has nothing to do with CEP, unless you can think of something.
I thought you were a native speaker.
E
Evgueni
New member
Dobro: Of course you've meant any recorder, and my answer to you shows that (I said about both digital and analog recorders).
When I wrote this:
> (Note that what dobro talked about was
> something different - he meant compression
> in the analog domain, before A/D)
I meant just the compression, which, as I figured out from the sequence you mentioned
(mic > pre > com > recorder) is still before ADC.
When I wrote this:
> (Note that what dobro talked about was
> something different - he meant compression
> in the analog domain, before A/D)
I meant just the compression, which, as I figured out from the sequence you mentioned
(mic > pre > com > recorder) is still before ADC.
camn
Active member
dobro.
My feild is reinforcement, but If I couldnt compress afterwards without loss of quality (which I can) Then I would record with compression. Ideally, I would have *some* amount, however small, on each percussive instrument, to counteract wild variations in strike velocity(unless my drummer was god)and because I totally prefer a nice punchy compressed kick. ALSO, on kick and snare you can eliminate crosstalk with a good compressor.
On strings I often use compression on bass and guitar, but before the mic.
xooxox
cameron
My feild is reinforcement, but If I couldnt compress afterwards without loss of quality (which I can) Then I would record with compression. Ideally, I would have *some* amount, however small, on each percussive instrument, to counteract wild variations in strike velocity(unless my drummer was god)and because I totally prefer a nice punchy compressed kick. ALSO, on kick and snare you can eliminate crosstalk with a good compressor.
On strings I often use compression on bass and guitar, but before the mic.
xooxox
cameron
P
PHATHEAD
New member
Here is a head scratcher'
if 32 bit is better than 24 bit simply on the premise that more is better, YOU ARE DEAD WRONG!!!!! not exactly ...... the truth of the matter is divisibility. Meaning ,if you know that everything you are recording will be in 32 or 16 bit word lengths you will be fine ,but if any source material has any dithering to do on a large scale 24 bit is better ,because 24 bit is divisible by 1,2,3,4,6,8,and 12 and 32 bit 1,2,4,8,and 16. wich means when it comes to dithering on the smaller numbers there will be less room for error and the small numbers are were you pick up the most distortion. This is all thoery and if you stay in the 32 bit realm all the time life is nice. especialy if you your end product in haeded to CD 16bit goes into 32 twice, that is it.
Phathead
if 32 bit is better than 24 bit simply on the premise that more is better, YOU ARE DEAD WRONG!!!!! not exactly ...... the truth of the matter is divisibility. Meaning ,if you know that everything you are recording will be in 32 or 16 bit word lengths you will be fine ,but if any source material has any dithering to do on a large scale 24 bit is better ,because 24 bit is divisible by 1,2,3,4,6,8,and 12 and 32 bit 1,2,4,8,and 16. wich means when it comes to dithering on the smaller numbers there will be less room for error and the small numbers are were you pick up the most distortion. This is all thoery and if you stay in the 32 bit realm all the time life is nice. especialy if you your end product in haeded to CD 16bit goes into 32 twice, that is it.
Phathead
E
Evgueni
New member
PHATHEAD: None of this has anything to do with divisibility of number of bits. I am afraid you've got your math confused. What you've just written makes no sense at all.
(I am sorry to write like this, but I am afraid someone will read that post and gets confused, so I had to do it
)
Just read what I've written above about the error accumulation, and you will see that every extra bit of resolution while procesing makes for less errors in the dithered down 16 bit result.
Which is, BTW, merely academic point, as 24 bits processing should already be more than enough for bringing the error accumulation rates to inaudible range (for reasonable amount of processing).
The computer-based workstations use 32 bits more as a matter of convenience and calculation speed (it's native word length for computer) than as a means to lessen the errors as compared to 24 bits results.
(I am sorry to write like this, but I am afraid someone will read that post and gets confused, so I had to do it
)Just read what I've written above about the error accumulation, and you will see that every extra bit of resolution while procesing makes for less errors in the dithered down 16 bit result.
Which is, BTW, merely academic point, as 24 bits processing should already be more than enough for bringing the error accumulation rates to inaudible range (for reasonable amount of processing).
The computer-based workstations use 32 bits more as a matter of convenience and calculation speed (it's native word length for computer) than as a means to lessen the errors as compared to 24 bits results.
dobro
Well-known member
So if the technology takes another jump up beyond 32-bit, that means there's no advantage for audio recording purposes?
E
Evgueni
New member
Dobro: Exactly.
And there is one more point to this. In the previois post I talked about 32 bits processing being much more than enough for 16 bits final output. But, as it happens, 32 bits is enough for processing for any final output resolution that can be reproduces by sound devices.
I've read a few times that any single resistor or interconnect adds amount of noise corresponding to the value of about 22nd bit of resolution at the line level. And now imagine how many of those (and other devices) are there in any single audio component.
So for the final product there is no point in having more than about 20 bits of _real_ resolution - just because no sound reproducing device on Earth is able to play it back without burying those small details in its own noise. This is just an estimation, and also the human ear can "listen into the noise" to some extent, but in any event the current "24 bits" standards are more than enough.
If your final result is going to have 24 bits of resolution (say, for DVD-Audio), then 32 bits processing adds 8 bits (32-24) margin - just enough for reducing error accumulation during any reasonable amount of processing.
So, one doesn't need more than 32 bits not only if the final result is going to be a 16 bits CD, but for any resolution that can be reproduced by the best possible sound devices.
And there is one more point to this. In the previois post I talked about 32 bits processing being much more than enough for 16 bits final output. But, as it happens, 32 bits is enough for processing for any final output resolution that can be reproduces by sound devices.
I've read a few times that any single resistor or interconnect adds amount of noise corresponding to the value of about 22nd bit of resolution at the line level. And now imagine how many of those (and other devices) are there in any single audio component.
So for the final product there is no point in having more than about 20 bits of _real_ resolution - just because no sound reproducing device on Earth is able to play it back without burying those small details in its own noise. This is just an estimation, and also the human ear can "listen into the noise" to some extent, but in any event the current "24 bits" standards are more than enough.
If your final result is going to have 24 bits of resolution (say, for DVD-Audio), then 32 bits processing adds 8 bits (32-24) margin - just enough for reducing error accumulation during any reasonable amount of processing.
So, one doesn't need more than 32 bits not only if the final result is going to be a 16 bits CD, but for any resolution that can be reproduced by the best possible sound devices.