Very unlikely I would say. This sound modulated noise is there indeed when not doing dithering, but! only when the LSB (bit23) is not random already. But even then, it would be still modulated noise with a peak value of the LSB.
Just do this test:
play the most annoing noise you can think of in the LSB only. (white noise, or for instance a square wave at 3KHz)
And add the most audible modulation you can thing of, I guess switching fully on and off second counts as very audible.
Play this at extreme max volume, for example 0dBFS == 120dB SPL.
The modulated noise would be ~-20dB SPL which is really inaudible.
The only way I can think of is that if Bob Katz can really hear a difference between 24 bit dither or not in a blind test this is caused by an other side effect of the dither/non-dither action.
And if you have something in the recording like the sound of a room at say -30dBA, is there any need of dither? I would think not!?
//
When you shift all bits one step to the right the LSB is lost and a zero is sucked in at the MSB position you achive a 50% (6,03dB) attenuation. Bit perfect actually - the ones still there anyway 
In my audio program Amadeus Pro for OS X if I do 6,03 or 6 dB it will not do arithmetics but if I ask it to do 50% it will indeed shift - says the designer... I had to verify it on .wav file level and it was so.
The in my DAC there will be wild calculations
//
In my audio program Amadeus Pro for OS X if I do 6,03 or 6 dB it will not do arithmetics but if I ask it to do 50% it will indeed shift - says the designer... I had to verify it on .wav file level and it was so.
The in my DAC there will be wild calculations
//
Are you referring to the special case where the signal is shifted by one bit while the output wordlength is longer than the input wordlength or in general, when you need to round off a signal?
For the general case, when you need to round off a signal and want the average and standard deviation of the total requantization error (round-off error including the effect of dither, if any) to be completely independent of the signal, in order to completely eliminate round-off-related distortion products and noise modulation, you have to apply 2 requantization steps peak-peak triangular probability density function dither.
When there is much more than a requantization step of noise in the signal anyway, there will hardly be any distortion or noise modulation when you don't dither. If you are perfectionistic you may still want to dither to completely get rid of requantization-related distortion and noise modulation, in fact that would be my personal preference, but it would be very unlikely that anyone ever hears any difference.
For 24 bit recordings it's usually a matter of perfectionism, but it's more important for 16 bit recordings with fade-outs played at a high volume. Besides, in a multibit sigma-delta DAC the signal may have to be rounded off to a noise-shaped five bits or so signal, then the behaviour really gets more predictable with dither.
Thanks! In all tests so far, even if I failed to put the files in the right order, I could hear the difference whatever that was... In the echo test I felt the biggest difference yet I failed completely. I though to play with that but it's truth I'm not sure myself about it.
IIUC dither used for gain changes is only needed before truncation. Therefore if 16-bit audio is to be played on a 24-bit player, no dither would be needed until the volume level is turned down so low that bit-15 of the 16-bit source (its LSB) would need to be shifted to below bit-23 (the new LSB) of the target 24-bit player. That's would be the point where a new truncation of the source data would need to occur.
Again IIUC, the level of dither is referenced to what will be the new LSB after truncation. Thus, 1-bit of dither of a 24-bit source would be applied to bit-15 if the audio is to be truncated down to 16-bits. OTOH, 1.5 bits of dither would apply noise at bit-15 and at bit-16.
Again IIUC, the level of dither is referenced to what will be the new LSB after truncation. Thus, 1-bit of dither of a 24-bit source would be applied to bit-15 if the audio is to be truncated down to 16-bits. OTOH, 1.5 bits of dither would apply noise at bit-15 and at bit-16.
Last edited:
Yes, please send Hans your 176.4 reaction report...
...as soon as possible. As Doede (dddac) indicates, it doesn't take long to evaluate and submit a report to Hans. Who knows, you may discover a method of better sounding digital playback of which you weren't previously aware
.
In DIY, we're forever questing for hidden musical treasure
.
...as soon as possible. As Doede (dddac) indicates, it doesn't take long to evaluate and submit a report to Hans. Who knows, you may discover a method of better sounding digital playback of which you weren't previously aware
. In DIY, we're forever questing for hidden musical treasure
.In DIY, we're forever questing for hidden musical treasure
If the scientific method is used, amatures like us can make an improvement.
Wow, that’s a thorough and carefull piece of documentation of a very interesting topic, chapeau.
Hans