HydrogenAudio

Original:
[attachimg=1]

Smart:
[attachimg=2]
[attachimg=3]

mda:
[attachimg=4]
[attachimg=5]

Audio files attached. Adding more bits won't eliminate the artifacts.

Thanks. I have always used Adobe Audition to verify correctness of the output and its frequency display doesn't seem accurate enough to show these issues. But I can see them with sox's spectrogram. The issue is in my output truncation function. I'll try to fix it soon.

Should be fixed in the just released version.

Yes fixed. Thanks!

Are these artifacts something to be concerned about for files processed previously to the fix? I am talking about 24 bit > 16 bit with 1 bit dither noise. Would you recommend to redo these or is this more of a cosmetic error which will be inaudible in practice?

No need to reprocess. The noise was only introduced with 23 and 24 bit output. And its loudness was under 130 dB below digital fullscale. I'd claim it was inaudible.

@Case, I applied 8-bit TPDF high-pass dither via MDA Dither (https://foobar.hyv.fi/?view=foo_dsp_mdadither) and Smart Dither (https://foobar.hyv.fi/?view=foo_dsp_dither) DSPs, which differ in how they handle near-silence gaps. Could you confirm that in the latter case the result is exactly what it should be, namely low noise in the first second, followed abruptly by a more noticeable noise? Or should there be no noise at all in the first second?

Spoiler (click to show/hide)

Good question. Smart dither component works as designed but this type of source signal combined with so few target bits is definitely not best for it. The beginning of the original track is dithered in a strange way, it contains several consecutive samples of silence combined with a sample or two of noise, then relatively long period of silence again. All the silence is retained by the component so it turns into a quiet background hiss. Once the pseudo silence ends the dither is constant so the background hiss gets louder.

If you dither that to a 16 bit target it will be flawless. No added noise and no background hiss.

This is a dilemma. It would be easy to filter too quiet source signal away into complete silence, but the point of dither is to retain as much of the original intented signal as possible.

@Case, can you think of a case when using 16-bit MDA Dither without taking silence into account is preferable to Smart Dither?

Only one thing comes to mind. There are DACs that turn themselves off when they don't receive input. If you have one of those you might want to prevent pure silence from entering the DAC during playback by having silence replaced by dither noise. That is the only reason I can think of that warrants replacing silence with noise.