Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: normalize volume with and without dither = should mixdown be silent? (Read 7735 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

normalize volume with and without dither = should mixdown be silent?

Hi,

I was testing out Adobe Audition 1.5 on a few various things, eg. running through same edits with and without dither to see if I could hear any differences etc., and came across something I'm not sure I understand...

I normalized an original track to -0.01dB, once with dither and another time without dither, then I overlayed the output files of the dithered and not dithered normalized files and made a mixdown of them in Adobe Audition 1.5, in the mixdown of the 2 tracks I can hear quietly the audio of the song with a very bassy bass drum... (this kind of surprised me as I expected it to be silent).

I then did the same thing with Audacity (instead of Adobe Audition), normalized a track to -0.01dB, once with dither and another without dither, when I made a mixdown it was silent, with NO music, unlike the Adobe Audition mixdown which had quiet music...

Is it odd that the same process in Audacity created a silent mixdown but the mixdown of the files created in Adobe had faint music?
Is Adobe Audition doing it's job properly there?

Thanks

normalize volume with and without dither = should mixdown be silent?

Reply #1 –
If Audacity gave you silence with and without dither then it was not doing what you think it was. This should never result in digital silence.

normalize volume with and without dither = should mixdown be silent?

Reply #2 –
Assuming you dithered at the 16th bit, i would not assume the dither to be easily audible on regular gear.

Generating full-scale sweep 16bits, no dither:
Code: [Select]
 sox -V -D -r 44100 -b 16 -c 1 -n  orig.wav synth 5 sine 300-3300


Reducing gain by 0.01dB, no dither added (the -D option):
Code: [Select]
sox -V -D orig.wav atten-nodither.wav gain -0.01


Reducing gain by 0.01dB, dither added (sox default):
Code: [Select]
sox -V orig.wav  atten-dither.wav gain -0.01


The mixdown (diff.wav):
Code: [Select]
sox -V -D -m atten-dither.wav -v -1 atten-nodither.wav diff.wav


The mixdown stats:
Code: [Select]
sox diff.wav -n stats
DC offset  -0.000000
Min level  -0.000031
Max level   0.000031
Pk lev dB     -90.31
RMS lev dB    -95.04
RMS Pk dB     -94.78
RMS Tr dB     -95.35
Crest factor    1.72
Flat factor     2.89
Pk count       74.3k
Bit-depth       1/16
Num samples     220k
Length s       5.000
Scale max   1.000000
Window s       0.050

 sox diff.wav -n stat
Samples read:            220500
Length (seconds):      5.000000
Scaled by:         2147483647.0
Maximum amplitude:     0.000031
Minimum amplitude:    -0.000031
Midline amplitude:    -0.000000
Mean    norm:          0.000010
Mean    amplitude:    -0.000000
RMS     amplitude:     0.000018
Maximum delta:         0.000061
Minimum delta:         0.000000
Mean    delta:         0.000017
RMS     delta:         0.000025
Rough   frequency:         9934
Volume adjustment:    32768.000


I cannot hear -90dB on my gear at standard listening volume setup.

normalize volume with and without dither = should mixdown be silent?

Reply #3 –
Quote
and made a mixdown of them in Adobe Audition 1.5, in the mixdown of the 2 tracks I can hear quietly the audio of the song with a very bassy bass drum... (this kind of surprised me as I expected it to be silent).
It's possible that Audition is doing some kind of "unexpected" dithering. I would expect different dithering algorithms with two different audio editors. The more "aggressive" the dither, the more likely you are to hear it in a null/subtraction test.

I assume it was VERY quiet and you had to crank-up the volume to hear it?

It's also possible that something's wrong with your "mixdown" procedure, which should of course be subtraction.

You've got two unknowns, so you probably also need to dither with one editor and subtract with the other to find-out where the difference is (2 more experiments).


You know dither is supposed to make a difference (an improvement), or there's no point in using it.    But in the real world at 16-bits, anything that's around 90 dB down is not going to make an audible difference (in a blind ABX test).  It's probably helpful at 8-bits (around 40dB down).

normalize volume with and without dither = should mixdown be silent?

Reply #4 –
In order for the result not to be the audio, at a higher volume, clipped on the peaks, you have to have inverted one of them. If the alignment were off by even one bit, the resultant would be far from silent, even with identical tracks.

normalize volume with and without dither = should mixdown be silent?

Reply #5 –
That could be the cause. Time misalignment of even a single sample between the two tracks produces result similar to that described by bomber1978

normalize volume with and without dither = should mixdown be silent?

Reply #6 –
Quote from: pdq on
If Audacity gave you silence with and without dither then it was not doing what you think it was. This should never result in digital silence.


With Audacity, when I say silence, there was actual hiss, from the dithering... But no music at all, not even faint.
So for example if with Audacity I changed a file speed by +1%, and did this with dither, then repeated the exact same step by taking the original file once again and giving it a +1% speed increase with dither I have created the same file twice.

I then create another 2 files from the original file, the exact same process of speeding it up +1%, but this time without dither instead of with dither.

So I now have 4 output files made individually from the original file, 2 with dither and 2 without dither, all 4 have a speed increase of +1%.

A mixdown of the 2 tracks without dither = total silence.
A mixdown of the 2 tracks with dither = hiss.
A mixdown of 1 track with dither and 1 track without dither = hiss.

I think what you are saying is that the mixdown of the track with dither and without dither should be producing some kind of audible "music"?

So it seems in the case of "speed change" editing Audacity dither simply adds hiss and does nothing else to the music!?

normalize volume with and without dither = should mixdown be silent?

Reply #7 –
Quote from: phofman on
Assuming you dithered at the 16th bit, i would not assume the dither to be easily audible on regular gear.

Generating full-scale sweep 16bits, no dither:
Code: [Select]
 sox -V -D -r 44100 -b 16 -c 1 -n  orig.wav synth 5 sine 300-3300


Reducing gain by 0.01dB, no dither added (the -D option):
Code: [Select]
sox -V -D orig.wav atten-nodither.wav gain -0.01


Reducing gain by 0.01dB, dither added (sox default):
Code: [Select]
sox -V orig.wav  atten-dither.wav gain -0.01


The mixdown (diff.wav):
Code: [Select]
sox -V -D -m atten-dither.wav -v -1 atten-nodither.wav diff.wav


The mixdown stats:
Code: [Select]
sox diff.wav -n stats
DC offset  -0.000000
Min level  -0.000031
Max level   0.000031
Pk lev dB     -90.31
RMS lev dB    -95.04
RMS Pk dB     -94.78
RMS Tr dB     -95.35
Crest factor    1.72
Flat factor     2.89
Pk count       74.3k
Bit-depth       1/16
Num samples     220k
Length s       5.000
Scale max   1.000000
Window s       0.050

 sox diff.wav -n stat
Samples read:            220500
Length (seconds):      5.000000
Scaled by:         2147483647.0
Maximum amplitude:     0.000031
Minimum amplitude:    -0.000031
Midline amplitude:    -0.000000
Mean    norm:          0.000010
Mean    amplitude:    -0.000000
RMS     amplitude:     0.000018
Maximum delta:         0.000061
Minimum delta:         0.000000
Mean    delta:         0.000017
RMS     delta:         0.000025
Rough   frequency:         9934
Volume adjustment:    32768.000


I cannot hear -90dB on my gear at standard listening volume setup.


I'm only listening working at my computer, with headphones and the volume half way up I can hear it as very low bassy music...

normalize volume with and without dither = should mixdown be silent?

Reply #8 –
Quote from: DVDdoug on
Quote
and made a mixdown of them in Adobe Audition 1.5, in the mixdown of the 2 tracks I can hear quietly the audio of the song with a very bassy bass drum... (this kind of surprised me as I expected it to be silent).
It's possible that Audition is doing some kind of "unexpected" dithering. I would expect different dithering algorithms with two different audio editors. The more "aggressive" the dither, the more likely you are to hear it in a null/subtraction test.

I assume it was VERY quiet and you had to crank-up the volume to hear it?

It's also possible that something's wrong with your "mixdown" procedure, which should of course be subtraction.

You've got two unknowns, so you probably also need to dither with one editor and subtract with the other to find-out where the difference is (2 more experiments).


You know dither is supposed to make a difference (an improvement), or there's no point in using it.    But in the real world at 16-bits, anything that's around 90 dB down is not going to make an audible difference (in a blind ABX test).  It's probably helpful at 8-bits (around 40dB down).


Yes it's very low, I'm working with earphones at my laptop, volume half way up it's very low bassy music.
I don't understand what you mean by do one test with the editor and subtract with the other?, can you please explain that?
Yes I know dither is supposed to make a difference..., makes me wonder if Audacity is completely useless at dithering, see my above post about the speed change test I did with and without dither...

normalize volume with and without dither = should mixdown be silent?

Reply #9 –
Quote from: AndyH-ha on
In order for the result not to be the audio, at a higher volume, clipped on the peaks, you have to have inverted one of them. If the alignment were off by even one bit, the resultant would be far from silent, even with identical tracks.


I'm not exactly sure what you are saying, but when I make the mixdown I do "Invert" one of the files before loading into multitrack.
I don't know if alignment works different in this case, but for example if I flip an original file left and right stereo 2 times back to the original way, and make a mixdown with the original file it's completely silent, so at least in that case the alignment is correct.., not sure if that changes though when doing a normalize volume edit....

normalize volume with and without dither = should mixdown be silent?

Reply #10 –
Kind of unrelated, but I'll put it here instead of a new thread.
I noticed with Adobe Audition if I edit a file by simply editing a 7min file down to 4mins with dither ticked on and another time ticked off both output files will not have the dither hiss at the silence before the start of the track - that seems normal to me as you don't need to dither when simpy trimming from 7mins to 4mins.
However if I "swap channels" it adds the dither noise in the silence before the track..., this is uneccessary isn't it? Not sure why an edit from 7mins to 4mins it knows not to dither, but when swapping channels it does not know that it doesn't need to dither!?!?!?!, and incorrectly does add dither...

**Another odd thing I find, is that if I speed change using Audacity, I can make a speed change about 10 times, and do an ABX test with the original un-edited file and I can only tell the difference about 2/10 times, however using Adobe Audition after I've made only a few speed changes in an ABX test I can tell the difference from the original pretty much 10/10 times, I think it seems most obvious by listening to the hi-hat... why would that be?, I always hear Adobe Audition is far superior to Audacity..........

normalize volume with and without dither = should mixdown be silent?

Reply #11 –
I do not think you should hear the original track faintly when subtracting dithered from nondithered. On a quality setup you could hear the added noise, as you did. IMO the most likely explanation for the audible result is that you shifted one of the tracks by one or more samples before the mixdown.

Quote
I always hear Adobe Audition is far superior to Audacity

What supporting arguments were the claims accompanied with to make them credible?

normalize volume with and without dither = should mixdown be silent?

Reply #12 –
Quote from: phofman on
I do not think you should hear the original track faintly when subtracting dithered from nondithered. On a quality setup you could hear the added noise, as you did. IMO the most likely explanation for the audible result is that you shifted one of the tracks by one or more samples before the mixdown.

Quote
I always hear Adobe Audition is far superior to Audacity

What supporting arguments were the claims accompanied with to make them credible?


When you say you think I have shifted one of the tracks by some samples before mixdown, is this something the programme/editing will do to itself?, because I certainly did not do anything manually, apart from add them into mixdown, and it creates the same result every time, so it's not a one off mistake I made...

I have noticed I get the same bassy low volume music mixdown when comparing a speed/pitch shift file of one with dither and one without dither compared to each other in mixdown.
However if I just make a cut/edit/trim of a file with and without dither the mixdown just gives silence/hiss without any music, also if I swap channels with and without dither, that mixdown will also only give silence without any music.
But as you have read a mixdown of something normalized/volume changed with and without dither does give the audible music in the mixdown, also a mixdown of a speed/pitch changed file with and without dither also gives the bassy low music mixdown - seems the 2 process' that genuinely require resampling create the audible music in the mixdown... But I'm not really sure if that's it...

normalize volume with and without dither = should mixdown be silent?

Reply #13 –
Quote from: bomber1978 on
I always hear Adobe Audition is far superior to Audacity..........
At least their marketing budget is far superior
I like your desire to find out what each DAW is really doing, and not rely on what others say. Been there, done that. Bugs and features are easily mixed up.
In my vocabulary subtracting one audio file from another is called "nulling", "null test" or "difference test".
It's strange indeed that you can still hear music, assuming Audition only added dither. If the two files aren't perfectly sync you should be able to hear music, but mostly high frequencies since LF will null more easily. A bassy sound of the difference signal might indicate another problem. There is a possibility that Audition creates a tiny bit of headroom to prevent overloads while adding (noise shaped) dither. You can easily verify this by redoing the test with audio normalized to say -6 dBFS.

About the "swapped channels" with dither: that's strange indeed, it should be lossless.

About the speed test: did you change speed + pitch (like sample rate conversion)? That's a relatively easy process that can be undone without significant artifacts. Changing tempo without changing pitch is very tricky, is likely to be audible and can't be undone without artifacts.

normalize volume with and without dither = should mixdown be silent?

Reply #14 –
Quote from: Kees de Visser on
Quote from: bomber1978 on
I always hear Adobe Audition is far superior to Audacity..........
At least their marketing budget is far superior
I like your desire to find out what each DAW is really doing, and not rely on what others say. Been there, done that. Bugs and features are easily mixed up.
In my vocabulary subtracting one audio file from another is called "nulling", "null test" or "difference test".
It's strange indeed that you can still hear music, assuming Audition only added dither. If the two files aren't perfectly sync you should be able to hear music, but mostly high frequencies since LF will null more easily. A bassy sound of the difference signal might indicate another problem. There is a possibility that Audition creates a tiny bit of headroom to prevent overloads while adding (noise shaped) dither. You can easily verify this by redoing the test with audio normalized to say -6 dBFS.

About the "swapped channels" with dither: that's strange indeed, it should be lossless.

About the speed test: did you change speed + pitch (like sample rate conversion)? That's a relatively easy process that can be undone without significant artifacts. Changing tempo without changing pitch is very tricky, is likely to be audible and can't be undone without artifacts.


Hi,

OK I'm not exactly sure I have done what you suggested correctly, but I have taken an original file and normalized to -6dB with dither, then took the original file again and normalized to -6dB without dither, I then made a mixdown of both the output files (-6dB with dither and -6dB without dither), and the faint bassy music is in the mixdown. I'm not sure if that is what you suggested I do so not sure if that sheds any light!?

Yes the "speed" test I did Adobe 1.5 vs Audacity 2.0.3 really was speed eg. changing both tempo and pitch.
The results have totally surprised me, I've always heard Adobe was far superior to Audacity. I changed the speed of an original file 10 times, with Adobe 1.5 and with Audacity 2.0.3, I also did a set for each of the 2 programmes with and without dither.
I put on my desktop 2 of the original file and 1 of the edited file, hid the labels and mixed them up so I had no idea which was which...
I then tried to pick the file which had been speed changed 10 times.
For Audacity I picked the file that had been speed changed 10 times only 2/10 times, for Adobe I picked the speed changed file 10/10 times.
I did a set dithered and another set not dithered, the second time I got the Audacity changed file right 3/10 times and again picked which file had been speed changed with Adobe 10/10 times.
So from what I can tell Audacity does a better job changing speed (tempo/pitch) than Adobe.
*Probably worth noting, this is with headphones and concentrating very carefully, I think it's most obvious when listening to a hi-hat, also this test was not done all 10 times on the same spot of music, I would move from 1min to 2min to 3min etc, yet picked the Adobe edited file 10/10 times, so I think it was a pretty reliable test I did...

Interestingly I then did this with a very old 1.2.6 version of Audacity from about 2007, and got pretty much exactly the same numbers...
*To be fair to Adobe though, when I only changed a file speed 2 times I don't think I could really tell the difference between it and an original file, so I guess it's OK because who really is going to change a file speed more than a few times?...

normalize volume with and without dither = should mixdown be silent?

Reply #15 –
OK I'm not sure but this might be proof that samples difference is the cause of audio in the mixdown???, I really need an expert opinion as I really don't know!

I took the original file and normalized to -6dB with dither, then took the original file and normalized to -6dB without dither (when played together in mixdown it gives a low bassy audio).

I loaded each file into Adobe 1.5 and went to analyze statistics, it shows there some info about samples (minimum sample value & maximum sample value), but I'm not really sure if this is exactly what you have been refering to?

It does indeed show 1 sample different for min + max sample value.

Here's the 2 screenshots.

Normalized -6dB with dither:


Normalized -6dB without dither: