Test your soundcard for clipping, with this sample Options

[Jan S.]

For everybody fighting with occasional clipping (or perfectionists wanting to test everything) you should try the attached sample, udial.

Testing:

• play the sample in your favourite player (decoded or not, having disabled all DSPs and EQs that can interfere)
  
• if the output sounds weird at any time you should:
  • lower the output vol. in your soundcard config (should be the speaker icon in your system tray).
  • For some soundcards 48kHz is better (audigy 2 etc.).
  • lower the vol. till the sample doesn't sound weird at any time.

The output will sound really weird if your settings clip.
Your output can occasional clip without you knowing/hearing it. But for us perfectionists this will theoretically give a better quality output.

ATTENTION: Play this sample at a low volume anytime, even if you hear nothing special! It can be very harmful to equipment and/or your ears.
It's strongly recommended to use some very cheap (PC) speakers if you want to test this,
otherwise you might really ruin your tweeters (it has happened several times already).

Attached File(s)

 udial.zip ( 217.55K ) Number of downloads: 19616

 

[DickD]

[EDIT]This post has some factual assumptions that were incorrect, as pointed out by Garf in the next post. I think I found the root cause in the post after his[/EDIT]

The FB2K database (if enabled) may have needed refreshing, I guess.

When I started typing this post, I suspected the console message is referring to clipping detected from the Resampler output. The tiniest amount of attenuation (volume control -0.5 dB) would be enough to stop this.

The 20 kHz +/- 1 kHz frequency-modulated sine wave in the original is almost full scale (probably a bit less so it can add to the extreme values of the dialling tones and reach full scale but no higher). The sampling points at 44.1 kHz are below full scale (hence the track peak of 0.997...) because they don't happen to coincide with the peaks.

Assuming I'm correct about the target amplitude of the sine waves, a perfect (infinitely long) reconstruction filter when upsampling should generate a track peak of 1.000000 at most on every peak where the sampling point lines up perfectly with the peak of the sinusoid. I'd imagine the Resampler (SSRC) DSP plugin, even in Slow mode (long reconstruction filter window) and 64-bit precision, will still show some very subtle (inaudibly subtle) variation in loudness at frequencies so near to the Nyquist limit (22.049... kHz is the highest frequency that can be accurately represented by 44100 Sa/s sampling). This could cause tiny amounts of clipping, such as 1.00001, which would generate inaudible clipping distortion, even on this sample, yet would be just high enough to trigger the console warning.

(I actually tried resampling to 48000, WAV 64-bit floating point to test this, and found that FB2K's resampler doesn't cause clipping on this sample).

Using Resampler: Slow mode, 64 bit:
Track Gain = +18.210000 dB
Track peak = 0.997059

Fast mode, 64 bit (same for 32 bit):
Track peak = 0.997217

So (without using Preamp or Equalizer) these shouldn't clip on their own.

What about adding dither? Could that cause clipping?

EDIT: No, it can't, as Garf points out below, the ditherer won't add dither if the added noise would push the sample value beyond +1.000 or -1.000.

Assuming you use 16-bit playback (which I'm not sure of), setting dither to "strong ATH noise shaping (recommended)" the stronger high frequency content than "no noise shaping" can add about 31 (out of 32767) to a sample value at the peak - just now and again. That's 0.000946, which still isn't enough to exceed 1.000000 if you apply no replaygain.

However, you're using ReplayGain, so it would like to add 18.21 dB, then realises it can't from the peak value of 0.997131, so it's scaling
to get a peak of 1.000000 instead (a gain of about +0.024955 dB). You then add dither, which can take the peak to 1.000946 at most).

Edit: This is not true. Dither is not added if the sample would exceed 1.000000, so it must already exceed 1.000000. The culprit is identified in my next message.

That is just enough to clip, but it's so incredibly subtle (and probably incredibly infrequent) that it clips, that it no audible clipping distortion is heard. However, you do get the warning from FB2K.

I guess it would be a finishing touch for FB2K's clipping prevention (Replaygain) to have an option to take account of the dither type and bit depth by scaling the peak to just below 1.000000 so that dither can't cause it to exceed 1.000000. We could either enter the margin by hand, or FB2K could have a look-up table for dither type versus peak dither amplitude for each bit depth and sampling rate. This is overkill at normal bit depths, where it's inaudible, but for low bit depths like 8-bit, it would be useful if people used strong ATH noise shaping.

Also, Replaygain on real music (this udial sample is NOT real music!) rarely causes clipping if the target volume is 89.0 dB - it hardly ever has positive Album Gain values, and only occasionally has positive Track Gain values.

EDIT: Please see my next post for the real explanation.

This post has been edited by DickD: Jun 3 2003, 11:42