32 Bit Float

[chrisdee]

Originally Posted by StewartFang

I'm telling you that when you record bed tracks 32 bit sessions sound different than 24 bit sessions which also sound different than 16 bit sessions.

Th reasons why are unimportant. They just do buddy.

When it comes to session sampling rates I can hear a improvement enough worth recording vocals at 96kHz compared to 44kHz. The recorded vocals sounds cleaner.
If I remember correctly someone explained to me it has to do with less aliasing at higher sampling rates.
So not the explenation I initially thought was logical (being higher samplings rates equals better/cleaner audio by itself).

However I don't hear any difference between 16, 24 or 32 bit sessions. Neither do I hear any improvement going from PT10.3.2 to PT10.3.3.

[Oblivion777]

Well I can tell from my personal experience that I and some of my colleagues did hear a difference between 48kHz and 96kHz recordings when mixing the material ITB (using lot's of different PI's obviously). Now whether that is a matter of the actual sample rate or the way different PI's are coded (how they handle the internal process of up/down sampling, etc.) I don't know ... nor care to be honest.

Regarding the initial subject of the thread: I always record 24-bit and have never made a blind-test to compare 16-bit vs. 24-bit vs. 32-bit float.

[Greg Malcangi]

Originally Posted by Oblivion777

Well I can tell from my personal experience that I and some of my colleagues did hear a difference between 48kHz and 96kHz recordings when mixing the material ITB (using lot's of different PI's obviously). Now whether that is a matter of the actual sample rate or the way different PI's are coded (how they handle the internal process of up/down sampling, etc.) I don't know ... nor care to be honest.

That's up to you of course but I personally do care about the quality of my work and therefore it's important to me to have a basic understanding of how and why things work so I can get the best quality possible out of the tools at my disposal for whatever delivery specification is required. But that's just me, each to his/her own.

There's ample evidence that human beings cannot hear the difference between 16/44.1 and higher bit depths/sampling rates of the same material but as you and some others don't care there doesn't seem much point discussing it.

G

[Oblivion777]

Originally Posted by Greg Malcangi

That's up to you of course but I personally do care about the quality of my work and therefore it's important to me to have a basic understanding of how and why things work so I can get the best quality possible out of the tools at my disposal for whatever delivery specification is required. G

Exactly; it's about the quality. The only difference from my perspective is that I don't need mathematical equations and precise laws of physics to explain every little detail. While I like theorizing from time to time and of course I really do not want to diminish the technicalities behind this subject I also like to use my ears as well. Anyway as you've written ... to each his/her own.

[Greg Malcangi]

Originally Posted by Oblivion777

Exactly; it's about the quality. The only difference from my perspective is that I don't need mathematical equations and precise laws of physics to explain every little detail. While I like theorizing from time to time and of course I really do not want to diminish the technicalities behind this subject I also like to use my ears as well. Anyway as you've written ... to each his/her own.

Where did I say don't use your ears? I said nothing about needing mathematical equations and precise laws of physics to explain every little detail, what I did say is a "basic understanding". Obviously your level of knowledge is such that you can't tell the difference between a basic layman's understanding and the complex math which underpins digital audio (which I haven't mentioned or quoted once).

So, the difference from my perspective is that what I (and everyone else) hears is an individual perception, I want my hearing perception to be as objective as possible by being reasonably well informed, you obviously feel differently ... to each his/her own, good luck with that! :)

G

[Drew Mazurek]

http://lavryengineering.com/pdfs/lav...ing-theory.pdf

I little light reading.

[dr_daw]

Originally Posted by dmazurek

http://lavryengineering.com/pdfs/lav...ing-theory.pdf

I little light reading.

That is a good read, I'm sorry for getting this thread off topic with Sample Rate. Now if we can get another link more clearly describing the theory behind 16 bit, 24bit and 32 bit.

I got lost in the hype there for a bit, and couldn't remember why I didn't use a higher sample rate than 48kHz. That paper by Lavry reminded me of what I've burned in brain cells over the years.

As for bit-rates, original topic, even if a 24bit chip only captures 20bits I would think that is better than a chip @ 16bits? as that would give you the headroom of 120dB (20bits) compared to 96dB, no? I mean I would rather a mixer with 120dB compared to 96dB of headroom. Is the theory the same with 24bit as with 32bit? Meaning is that last 4 bits of a 20bit ADC an exponant? Does it not bring the noise floor to a lower level, causing the quantization errors to be inaudible?

I mean it's been a long time since I recorded in 16bit, but I'm possitive that things sounded a bit better at 24bit. Placebo effect?

[CME]

Originally Posted by dr_daw

That is a good read, I'm sorry for getting this thread off topic with Sample Rate. Now if we can get another link more clearly describing the theory behind 16 bit, 24bit and 32 bit.

I got lost in the hype there for a bit, and couldn't remember why I didn't use a higher sample rate than 48kHz. That paper by Lavry reminded me of what I've burned in brain cells over the years.

As for bit-rates, original topic, even if a 24bit chip only captures 20bits I would think that is better than a chip @ 16bits? as that would give you the headroom of 120dB (20bits) compared to 96dB, no? I mean I would rather a mixer with 120dB compared to 96dB of headroom. Is the theory the same with 24bit as with 32bit? Meaning is that last 4 bits of a 20bit ADC an exponant? Does it not bring the noise floor to a lower level, causing the quantization errors to be inaudible?

I mean it's been a long time since I recorded in 16bit, but I'm possitive that things sounded a bit better at 24bit. Placebo effect?

The problem is there are no converters that are exporting 32-bit words. So the last bits are empty if you are recording through a 24-bit converter into 32-bit float files.

Now once you've recorded the files there are advantages to changing the session to 32-bit float.

And Stewart. I haven't asked. Are you mostly recording VI's in the box? If so there may be advantages in 32-bit float. I'm not sure, but VI's may be able to output true 32-bit float audio. That's something I'm not familiar with.

And as far as 24-bit vs.16 bit, you do move the noise floor lower and this does add up to less noise as you start summing tracks.

However who is recording sounds with these huge dynamic (100+db) ranges? And what equipment are you using to do so? I'm not opposed to higher bit depth recordings. Just don't think there is anybody out there capable if pushing the limits of 24-bit.

[Greg Malcangi]

Originally Posted by dr_daw

That is a good read, I'm sorry for getting this thread off topic with Sample Rate. Now if we can get another link more clearly describing the theory behind 16 bit, 24bit and 32 bit.

I'll try to explain a little. For a better explanation, Digital Audio Explained is a book anyone serious about audio engineering should know.

Because bits are binary, every time we add a bit of digital data we double the number of available values, so 4bit has 8 values, 5 bit has 16 values, 6 bit 32 values, etc. Each time we double the number of available values we statistically halve the number of quantisation errors and quantisation errors end up (due to the action of the dithering quantiser) as white noise. So each additional data bit reduces the noise floor by half, which is -6dB. This is how we arrive at 1bit = 6dB more headroom. So, the added resolution of a higher bit depth results ONLY in a lower noise floor, there is no other additional improvement in fidelity. In other words, we can record a signal just as perfectly with 1bit as we can with 24bits, just that the 1bit recording will be swamped with massive amounts noise. The proof of this is the existence of the SACD format, which uses just 1bit of data (all the resultant noise is moved into ultrasonic frequencies where it can't be heard) yet appears to have at least as high a fidelity as 24bit recordings.

In theory then, provided it is properly gain staged, a 16bit recording cannot sound inferior to a 24bit recording. In practice we can't say for sure that an ADC manufacturer has implemented 16bit as well as 24bit recording but it's very unlikely an ADC will have a design flaw for only it's 16bit setting.

Originally Posted by dr_daw

I mean I would rather a mixer with 120dB compared to 96dB of headroom.

Now you're talking about something quite different. The dynamic range of an entirely digital summing buss is a different proposition to an ADC or DAC which have to deal with the physics limitations of the analogue world. The summing mixer in ProTools TDM is 48bit fixed which provides 288dB of SNR and in other versions of PT the summing mixer is 64bit float so again, a huge range.

Originally Posted by dr_daw

Is the theory the same with 24bit as with 32bit? Meaning is that last 4 bits of a 20bit ADC an exponant? Does it not bring the noise floor to a lower level, causing the quantization errors to be inaudible?

No, 24bit is fixed point as opposed to 32bit floating point, which means with 24bit there is no exponent, just 24bits of data or rather, 20bits of signal data and 4 bits of noise. 32bit float contains the same 24bits of data (20bits signal, 4bits of noise for example) as a 24bit fixed point file, no more and no less, the additional 8bits are reserved for defining where the floating point is (the value of the exponent). Regardless of how many bits of data you use to record an audio signal the ultimate limitation is the laws of physics. You need to appreciate the enormous dynamic range that 24bits of data represents (144dB). Just the sound produced by the electrons colliding within a 1.5k resistor equates to about -138dB. In other words, if the analogue input circuitry in your ADC only contained a single 1.5k resistor, your ADC would only be able to resolve 23bits of audio data, the 24th bit would only contain the sound of electrons colliding in our resistor. Of course there are more components than a single 1.5k resistor in the analogue input circuitry of an ADC and therefore more noise. It is this noise which defines our theoretical recording limit of about 126dB and is never going to change unless we find a way of stopping electrons colliding and AFAIK that would require a change to the laws of physics! This 126dB limit is purely theoretical and cannot be achieved in practice and is even more unobtainable in a real world recording situation. Forget about the ridiculously low level sound of electrons colliding in your ADC, what about the noise floor of your recording environment and the internal noise of your mic and mic-pre? Summing all this noise, even a world class live room, mic and mic-pre is going to struggle to achieve 100dB of SNR, making the dynamic range of our effectively 20bit ADC more than sufficient.

One last point, there are virtually no commercially released recordings which have a dynamic range of more than about 60dB, that of course equates to about 10bits, making even 16bit complete overkill as a consumer delivery format.

Hope this helps?

G

[Greg Malcangi]

Originally Posted by StewartFang

You guys basically have no clue and should really educate yourselves properly before talking about something...

This has got to be the finest example of hypocrisy I've ever had the misfortune to read, you must feel very proud of yourself!