32 bit float to 16 bit PCM conversion process

I have a wave file open in a hex editor to get a good idea as to what’s going on with the data, I have an all zeroes section that allows me to see how values are displayed in Audacity. The values in the displayed 32 bit float format are:

01 00 02 00 03 00 04 00 05 00 06 00 07 00 08 00 
09 00 0A 00 0B 00 0C 00 0D 00 0E 00 0F 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

When I save that file as 16 bit PCM and open it to compare it with the 32 bit float I see that memory region as:

01 00 01 00 05 00 01 00 08 00 04 00 07 00 09 00
09 00 09 00 0C 00 0C 00 0B 00 12 00 0A 00 04 00
FF FF FF FF 01 00 FF FF 01 00 00 00 00 00 FF FF
02 00 FE FF 01 00 00 00 00 00 00 00 00 00 00 00

Can someone point me to the Audacity file that converts 32 bit float to 16 bit PCM? I understand that the down conversion would result in value changes due to the lower bit count, but not what I’m seeing, especially when the values are zero. Also, I do not expect the down conversion to result in the “jagged” values, such as from 3 to 5 and 4 to 1. Audibly, these mean nothing, but I’d just like to understand what’s going on in the code.

Thanks in advance.

This is probably what’s causing the confusion: https://manual.audacityteam.org/man/dither.html

If you turn off dither, then you’ll probably see the values that you are expecting.

I’ve never looked at the source code and I don’t know how the bytes are arranged with floating-point WAVs…

But maybe this will help - The 0dB reference for regular (integer) WAV files is the maximum count (+32,767 or 32,768 with 16-bits). In floating point 0dB is represented by +/-1.0.

I do not expect the down conversion to result in the “jagged” values, such as from 3 to 5 and 4 to 1.

Are you talking about [u]dither[/u]?

Thanks for your replies. I didn’t know about dither, now I do. However, Audacity’s conversion, Dither set to None for High-quality Conversion, didn’t result in all zeroes where the original 32 bit float signal had zeroes:

01 00 02 00 03 00 03 00 05 00 05 00 07 00 07 00
09 00 09 00 0A 00 0B 00 0B 00 0E 00 0C 00 11 00
06 00 FE FF 01 00 FF FF 01 00 FF FF 01 00 FE FF
02 00 FF FF 01 00 00 00 FF FF 03 00 FC FF 04 00

But now I know what to expect, a little fluctuation around higher order “zero” values at conversion.

It does for me:

16-bit silence.png911×627 44 KB

32-bit-float.png911×627 44.2 KB

OK. I did this after I booted my computer from scratch and it appears that my issue is no longer happening. I tried the same process again and see no fluctuations, all values in the 16 bit PCM file in the “zero region” are the same as those in the 32 bit version. It must have been the state of Audacity that caused it.