Recording at 24bit appears to be 16 bits padded to 24bit

[steve]

Notice the difference in the positions of the individual samples in a 32 bit track compared to a 16 bit track:

Upper track is 32 bit, lower track is 16 bit

screenshot.png (6.82 KiB) Viewed 2686 times

[thunderstick]

Well, I have no idea how you achieved those graphical results.

EDIT: Ok, I didn't use the zoom feature itself for this test, now I can replicate what you've posted.

With regards to this thread, I wonder if infact a 16bit file padded to 24bit would show samples below -96dB?

What does it technically mean when a 16bit file is padded to 24bit? Is it that the file itself is recorded with 16 bits per sample but the file header has been tagged as 24 bits per sample, and thus any program reading the file reports the file as such?

Does this shot show a value below -100dB?

[steve]

I wonder if infact a 16bit file padded to 24bit would show samples below -96dB?
What does it technically mean when a 16bit file is padded to 24bit?

If you record from a 16 bit device then each sample is in the range of 0000000000000000 (0 in decimal) and 1111111111111111 (131071 in decimal). Each sample will appear in an Audacity track with a value in the range of +1.0 to -1.0, so basically that means that there are 131072 steps between +1.0 and -1.0.

in 24 or 32 bit, there are many more steps, so taking as an example the pictures in my last post, a very low amplitude sine wave can be represented as a smooth curve, whereas a 16 bit representation shows noticeable steps.

Digital audio is always an approximation as the actual sound wave is analogue - there are an infinite number of steps - but in theory, as the bit depth increases (the available steps are made smaller), then the approximation to the analogue signal becomes more accurate. In practice 16 bits gives a very good approximation, and increasing the bit depth beyond 16 gives only very marginal improvement in sound quality.

If we record a 16 bit device as a 24 or 32 bit track, then each sample value will be represented as a value between +1.0 and -1.0 in 16 bit resolution. For example, a sample value sent from the sound card may be 0000000000010000 or 0000000000010001 but it can not be half way between these values. In a 32 bit format, there are thousands of intermediate steps between these values, but they will not be used. Only 16 bit values will be used.

In the original enquiry, 24bit hardware was being used, but Audacity was not receiving 24 bit data from it, just 16 bit data, which I can only presume is because the sound card driver is running as 16 bit.

We sometimes see issues where sound cards do not work as well as they should in Audacity because Audacity, as open source software, is distributed with support for the operating systems' sound system, which for Windows machines means that it will use the native Windows (WDM) drivers. It does not support third party closed source drivers (notably ASIO drivers are not supported) because it is restricted from doing so by licensing terms. Sound card manufacturers of high quality sound cards will generally release several different drivers, including WDM drivers, but in this case it seems like they were lazy and only provide 16 bit versions of the WDM drivers.

How much does this matter? Not much (in my opinion).
As explained, digital audio is representing a continuously variable (analogue) signal in discrete (digital) steps, but the sound card does not output the sound as a signal that jumps from one digital step to the next, the output is filtered to produce a smooth, analogue style continuously variable output. So the difference in the sound of a 16 bit wave and a 32bit wave will be virtually undetectable. The most noticeable difference between the two (and even his is not very noticeable) is with very low level sounds. For 16 bit sound, the first digital step is between silence (minus infinity dB) and -96 dB, so any sound in this range must take one of these values and cannot have a value in between. This sets the noise floor (at best) at -96 dB. With 24bit audio, the noise floor can theoretically be much lower, but in practice noise from elsewhere will take over as the limiting factor. If you connect a microphone to your sound card the noise level is likely to be much louder than -96 dB due to self noise of the microphone and the analogue circuitry in the microphone pre amp (and add onto that any noise present in the recording environment such as noise from your computer fan, nearby traffic, the TV next door...)

Even if a 16 bit sound card is being used, it can still be worth recording in 32 bit as the increased resolution makes the numerical calculations that are involved in processing the sound far more accurate, so you can do a lot of processing with virtually no loss of sound quality. If you are just recording something and burning it to CD (16 bit), then there are arguably no advantages in using 32 bit.

Does this shot show a value below -100dB?

Yes. You can see them clearly in the lower channel.

[eharmonica]

Thunderstick:

yes those samples look fine.


Steve: I am using the latest drivers from Emu. Also, I upgraded to the latest BEtA version od Audacity, and now I can't even select the Emu as a recording device. Ugh! Additionally, it makes me have the playback device the same as the recording device.

As for 16bit vs 24bit soundwise: it's not that one is significantly better, it's that when I record something and share it with my crowd and I tell them it's 24-96, but then it turns out that it is NOT, then I look bad.

[steve]

I upgraded to the latest BEtA version od Audacity, and now I can't even select the Emu as a recording device. Ugh! Additionally, it makes me have the playback device the same as the recording device.

What options do you have for the recording and playback devices?
This is with Audacity 1.3.7 on Windows XP?

By the way - do you still get 16 bit resolution in Audacity 1.2.6 if you set Audacity Preferences (Quality tab) to 32 bit rather than 24 bit?

[thunderstick]

I wonder if infact a 16bit file padded to 24bit would show samples below -96dB?
What does it technically mean when a 16bit file is padded to 24bit?

If you record .......... then there are arguably no advantages in using 32 bit.

Does this shot show a value below -100dB?

Yes. You can see them clearly in the lower channel.

Thank you for the tutorial.

Thunderstick:

yes those samples look fine.

Thank you for the analysis. I hope you can find a version that works for you, as I realise (and know) how frustrating these things can be. The good thing is at least CEP seems to work for you, so that in itself is a relief.

[eharmonica]

Field as many outstanding inquiries as I can remember:
I'm now using Audacity 1.3.7 on a Vista Dell laptop. My internal soundcard records on CEP at 24bit, Audacity (1.2.6) gives me 16bit.
I must confess that I am having trouble figuring out how to export in 24- or 32bit in 1.3.7. It defaults to 16bit, and I have to select "Other" to get it out. I'm not much liking 1.3.7.

The Emu selection issue went away with a reboot

Whatever the settings, 24- or 32bit, I get 16bit results.

I will check out Audacity on my XP laptop when I get the chance

[thunderstick]

I must confess that I am having trouble figuring out how to export in 24- or 32bit in 1.3.7. It defaults to 16bit, and I have to select "Other" to get it out. I'm not much liking 1.3.7.

When you choose to export, there should be a window that pops up where you select the file name and location. There should be an options radio button in this pop-up window. That is where all of the export settings are found. That's the way it is in 1.3.3, anyhow.

[thunderstick]

Another thing you should do is delete the preferences registry entry when you switch from 1.2 to 1.3.X. HKEY_CURRENT_USERSoftwareAudacity

The preferences settings for 1.3.X are stored in Documents and Settings<user name>Application DataAudacity

[Gale Andrews]

The "DirectSound" choice for SoundBlaster that Thunderstick is using in the Audacity Audio I/O Preferences might be the key here, as the other choice (MME) only supports 16-bit recording (even with 24-bit hardware). Steve has told me he thinks CoolEdit uses DirectSound, but 1.3.7 is the first Audacity version where we have published our DirectSound support.

So Enharmonica, if your EMU drivers let you record in CoolEdit, it should be possible to record in true 24-bit resolution in Audacity 1.3.7 if you use the DirectSound choice, and select 24-bit (or 32-bit) in the Default Sample Format in the Audacity Quality Preferences. You must use the DirectSound API for both playback and recording, though.


Gale