http://forum.image-line.com/viewtopic.php?t=6798
I wonder how the 24-bit files are truncated to 16-bit in Sampler/Audioclip Channels? I know that 32-bit floating point files ARE NOT truncated, yet the 24-bit integer files ARE.
Tests I made are quite simple and show some weird results:
1) I took a 24-bit file and inserted it in Audioclip (as to not match volume),
2) rendered it to 32-bit floating point (further as 32-bit FP),
I also made another version of the test with additional step:
2b) truncated the rendered file to 24-bit integer file in Sound Forge.
3) then I subtracted the rendered file from the original in Sound Forge (i also did it in WaveLab to exclude audioeditor-specific- behaviour/bugs).
If the files were identical, the resulting file should be pure digital silence (all samples have value 0).
However that was not the case, when 24-bit files were taken as originals.
The subtraction resulting file consised of digital virtually 1-bit noise - samples had values either equal 0, or equal (only in positive scale area) 2111(in 32-bit FP) / 256 (in 24-bit integer) / -90,3 dBFS) . And it IS NOT what the simple truncation would result in. The subtraction result of a 24-bit file and its copy derived from truncating the original to 16-bit is not 1-bit by the charcter, and its values are evenly spread in the -128... +128 for 24-bit /-15259...+15259 for 32-bit FP range, which makes - 96,3 dBFS. So, it feels like something messes up 15-th bit in the original file, so its value either coincides with the original file, or not.
Please explain what happens to a 24-bit integer files in the Sampler/Audioclip Channels.
Original 32-bit FP files tested in the same way result in 0 - pure silence.
I wonder how the 24-bit files are truncated to 16-bit in Sampler/Audioclip Channels? I know that 32-bit floating point files ARE NOT truncated, yet the 24-bit integer files ARE.
Tests I made are quite simple and show some weird results:
1) I took a 24-bit file and inserted it in Audioclip (as to not match volume),
2) rendered it to 32-bit floating point (further as 32-bit FP),
I also made another version of the test with additional step:
2b) truncated the rendered file to 24-bit integer file in Sound Forge.
3) then I subtracted the rendered file from the original in Sound Forge (i also did it in WaveLab to exclude audioeditor-specific- behaviour/bugs).
If the files were identical, the resulting file should be pure digital silence (all samples have value 0).
However that was not the case, when 24-bit files were taken as originals.
The subtraction resulting file consised of digital virtually 1-bit noise - samples had values either equal 0, or equal (only in positive scale area) 2111(in 32-bit FP) / 256 (in 24-bit integer) / -90,3 dBFS) . And it IS NOT what the simple truncation would result in. The subtraction result of a 24-bit file and its copy derived from truncating the original to 16-bit is not 1-bit by the charcter, and its values are evenly spread in the -128... +128 for 24-bit /-15259...+15259 for 32-bit FP range, which makes - 96,3 dBFS. So, it feels like something messes up 15-th bit in the original file, so its value either coincides with the original file, or not.
Please explain what happens to a 24-bit integer files in the Sampler/Audioclip Channels.
Original 32-bit FP files tested in the same way result in 0 - pure silence.
