Audacity (2.2.1 on Mac OS 10.12.6) adds noise above 13 kHz (and between 8 and 11 kHz) whenever I export an audio file. Attached are two files: Test.aif, which I opened with Audacity, and Test-export.aiff, which is the otherwise unchanged result of Audacity’s export function. The noise can be seen in any spectrogram with propper settings, and it can be heard clearly when played back slower.
Audacity adds a dither signal to outgoing WAV, AIFF and probably others unless you stop it. It helps mask downconverting distortion from Audacity’s super high quality internal format to “normal” uncompressed formats.
You can stop it from happening in Preferences, however, you might get standards errors unless you use 32-bit floating for your export.
People who find that signal are usually scientists or techies trying to use Audacity for critical science experiments or testing. Audacity is an audio production editor, not a wav editor and that use is probably not a good idea.
Audacity goes through two conversions for every show, once going in and once again going out.
I bet you’re wondering why on earth would we do that?
For one example, the recommended audiobook mastering protocol sometimes causes intentional peak distortion which is recoverable to clear sound because of this double conversion process.
What spectrogram settings? Could you post the plots?
“Can be heard when played back slower” - but can’t be heard when played back at normal speed? If so, what’s the problem?
My guess is that you are referring to “dither” noise. See here for information: https://manual.audacityteam.org/man/dither.html
Yes, Thanks, Koz and steve, I stopped the dither and the result is fine. I use Audacity only because it is very convenient to do simple editing (e.g. it can open multichannel sound files), I don’t use any of the effects. Maybe no dither by default would make more sense?
@Bill: I’ve attached the settings for an Audacity spectrogram (in German). I wrote the noise can be heard clearly when played back slower, but it can also be heard in normal speed, depending on the content of the file (try with silence!). The noise is definitely a problem if you use the exported sound file for any further processing.
but it can also be heard in normal speed
I wonder if something else is broken. The default dithering signal is not audible in normal production.
Post a screen shot of what your normal sound looks like on the timeline.
This is approximately correct.
If you are recording and producing with very, very low volume, your show and the dither signal may be too close to each other in loudness. If you like the Waveform dB setting on the timeline, you may not be able to tell there’s anything wrong.
Are you using the right dither?
Maybe no dither by default would make more sense?
Probably not. Opening sound in Audacity is “free” because the quality inside Audacity is almost always better than the sound file.
But exporting can have problems. You can have much more detail in the Audacity sound than the export format will support. Unfortunately, the conversion errors can be serious. Make up new sound that didn’t used to be there, invent fake pitch changes or drop sounds that exist.
Small amounts of dither keep the errors from lining up and creating problems. The trick is to make the dither signal not create other problems.
was an answer to Bill. It did not refer to normal production circumstances, hence the reference to silence.
As I wrote, the noise creates problems when the exported file is used further. In short: I use Audacity as an editing tool for sound files which I’ve recorded in Max/MSP and which I then use again to simulate the performance situation and further develop my Max patch. Audacity is very convenient for this kind of editing. Maybe I’m the only person using it in this way, but I don’t understand why there is dither in this case – there should automatically by default be no dither when a file is being exported with simple editing only and without any processing.
Can’t you hear the white noise at the end of the “Test.aif” file?
Sorry, I don’t understand why you ask this. Of course I can hear it, but that’s not the noise I was talking about. Test.aif has not been exported by Audacity, I was talking about the noise (dither, as you rightly guessed) which is added by Audacity in Test-export.aiff (even that can be heard of course when amplified, but more clearly and without amplification when played back slower).
I don’t understand why you are complaining about the extremely low level noise in Test-export.aiff but not complaining about the much louder noise that was already present in Test.aif.
16-bit PCM audio will always (except for absolute silence) have a base level of noise of around -90 dB because all sample values are integers. It’s called “quantization noise”. When done correctly, dither replaces quantization noise with noise that is shaped to be less audible than un-dithered quantization noise. If you prefer un-dithered quantization noise, then you can turn off dither in Audacity’s preferences. Standard practice in professional audio production is to apply dither when down-sampling to 16-bit. It’s also standard practice in professional audio production to not down-sample to 16-bit until all editing and processing is completed.
The already present noise is not louder, as can be seen in the spectrograms attached (for the settings see my earlier posting). Audacity is, at least in this case, not replacing audible quantization noise by less audible dither noise, but simply adding noise (which may, I admit, in many cases not disturb). All I suggest is that dither should not be applied when it doesn’t serve any purpose.
Thanks for your explanation anyway. I’m sure the default settings are perfect for many users.
I disagree. With my ears I clearly hear a background “shhhh” at the end of the Test.aif file (in addition to the “eeee”). At the end of the Test-export.aiff file I hear (with my ears) the “eeee” and the “shhhh”. If I turn the volume up very loud (which would make the start painfully loud), then I can hear that the “shhhh” has a sharper timbre, but I would not normally turn the volume up that loud because I value my hearing.
Where does the Test.aif file come from and how do you know that it doesn’t have rectangular dither?
I did exactely the same before (for testing purposes) and I heard the same: the shhhh is slightly sharper. The reason why the dither noise doesn’t have a stronger impact is simply because it is very high. When you play the file slower, i.e. transpose the sound, the dither noise becomes quite loud. As I mentioned earlier I use edited, exported files for testing purposes in Max/Msp, and the processing there involves transposition.
The Test.aif file was recording with Max/Msp, I’ve never heard about dither in Max.
If you compare the two spectrograms you can see that there is much more high pitched noise in Test-export.aif at the end, and at the beginning the spectrum seems to be altered slightly (although I find it difficult to see). Again: I think the spectrum and the sound should not change at all when only simple editing is involved.
Sure, but if you want the track slower you should slow it down before applying dither. Similarly, if you want to boost the treble, you should do that before applying dither or the dither noise will become noticeable.
I agree, but:
- In the vast majority of cases, people also apply some kind of processing, such as normalizing, fading …
- Determining whether sample values have changed during the course of working on a project is non-trivial. Audacity can’t rely on the project “History” because the project may have been saved and re-opened, which loses the project history from the first session.
My preferred solution to this problem is to make the dither option more easily available, such as a check box (enabled by default) in the Export dialog.
Ok, I understand. Your preferred solution sounds very reasonable to me. If it ever gets implemented I would also suggest that Audacity remembers unchecking.
Thanks for your time.