bit depth - converting to a higher depth

BigKahuna

New member
This is not a computer question ... but it concerns digital audio, which I think fits in at this forum.

I have a bunch of tracks that were recorded at 20bit/48kHz onto ADAT ...
I want to transfer them onto a HD recorder (through lightpipe) ...
can I get away with just recording them at 24bit/48kHz ... or do I have to keep them at 20 bit to avoid any glitches in the audio?
If I can transfer them at the higher depth, do I need to dither (or ... ummm ... I guess "UNdither" in this case)?
I've transferred one song and it sounds OK to me ... but I'm wondering if there's a rule I should follow.
Thanks much
 
I dont know how lightpipe/ADAT works, but Ive taken 16 bit files and recorded them as 24 bit, no problems......
 
Really.......thats interesting. You mean bits may added to an existing 16bit file? Hmmmmm. I need MORE KNOWLEDGE ON THIS SHIT!!!!!!!!!!!!
 
My understanding's this: you can go up in bits from 16 to 24, or from 20 to 24, with no degradation in sound. There will simply be 4 unoccupied bits in every sample. No glitches, no crap in the sound.
 
thanks guys ... that's kinda what I thought .. because I didn't notice any problems with this one song I recorded. Unless anyone can point out any problems they've had converting to more bits ... I'm gonna run with it and transfer the rest of the songs to the HD recorder. Yee haw!
Thanks again for your input!
 
I will hazard a guess as to why this is possible.

audio is digitally stored on DAT and goes therough lightpipe. The soundcard converts it from one format of digital information either into audio or into some floating point digital info format and then the sound is re-recorded as if from new to the new Bit depth etc.

As I said a guess....
 
Actually it's very simple.

Let's just look at an individual sample peaking at the very top of the waveform (right at 0dB) from a mono 16 bit file. I would look like this:

1111111111111111

This 16 bit sample has approximately 96dB signal to noise ratio.

To convert the sample to 24 bit the software just puts these in the top 16 most significant bits and set the rest as zeros. Like this:

000000001111111111111111

That's all there is to it. Now, 24 bit files can potentially have about 140dB signal to noise, but this file still only has 96dB. Those place holder 0's at the beginning are essentially noise down at -96dB.

These 0's can present a problem for some DSP algorithms. They can be potential sources of aliasing distortions. In most cases it has no audible effect since it's happening below that -96dB noise floor. But for high resolution mixes it's good to have an added measure of security by filling those lowest bits with random noise. This is akin to BigKahuna's "UNdither". But of course you're not gaining information, you're just changing the nature of the place holders.

All you do is take a full scale white (or pink) noise 24 bit wave file and lower the volume to -96dB, then mix it in with your converted file. This can help smooth out potential low level processing errors. You don't sacrifice anything because your original 16 bit file was only 96dB signal to noise anyhow.

barefoot
 
Last edited:
Back
Top