Sample rate conversion and dither tests.

DISCLAIMER:
I did this test just to hear for myself whether it makes that much difference what route to take out of Pro Tools to do the best
samplerate-conversion and what dither to apply.
I am in no way a scientist, nor am I claiming to know what is best for your audio.
Last but not least, I don’t want to pit one brand against the other. Probably there are dozens of better algorithms out there or
plugins that out-perform all of the ones mentioned here. It is just what is in my toolbox and at my disposal.

To be able to perform tests yourself, you can download the Pro Tools sessions here: https://infinit.io/_/tWRwDFw

The scenario is simple: I have a session of a radio documentary, delivered by the client in 32bit 44.1kHz. After editing and
mixing, the master needs to be converted to 16bit 48kHz to be uploaded to a broadcast-server that will eventually convert it to
MPEG1 Layer II (mp2) for final ingestion for play-out.

For practical purposes I choose not to use the full 40’ doc, but made a small test file to work with, consisting of:
- Fragment of the radio documentary (-23 LUFS)
- 1000Hz Sine, Square, Saw, triangle, Pink and White noise (all at -20dB)
- Full Sweep -6.00 dB, 24bit 44.1 kHz, 10 sec
- Snippet of over-compressed hip-hop from an MP4

I imported the different results in a 32bit 48kHz, flipped the phase of the original to be able to check the differences, which I
pulled up by +36dB.
In the session-screenshot on the next page, you can see I immediately noticed some abnormalities when working with iZotope’s
RX, which is considered by many to do the best job.
Flavors of dither by ear:
It is obvious not much is happening when exporting without SR-conversion. There’s dither but we’ll come to that later.
But as soon as we do SR Conversions you can see that the over-compressed hip-hop (probably with inter-sample peaks) is
showing. But with RX (used both through RX Connect as standalone) much more is happening…. Take a look at the 3rd track
from below. This is where I did resampling first, followed by dither and than exported the file, without dither.
This is what it looks like in RX:
It clearly struggles with White and Pink Noise (distortion on the meters) and leaves a nasty blip at the end of the sweep.
When set to 0dB the White Noise measures a whopping -27.9 dB against an average of -75dB with the other files. Not good.
Also AudioEase’s Barbabatch, of which I always was an avid fan, struggles: -40dB on the White Noise when the clip is set to -0dB
and the blip at the end of the sweep also shows. It looks like this:
Dither
I first listened to the sound of the dither itself on closed headphones. And this is what I noticed, which clearly is not always what
is measured.

- P r o T o o s D i t h e r is a high hiss. I could not differentiate between dither on the track itself, on a master with bouncing or the
automatic dither on export.

!
- P O W - r sounds lower, almost like tape hiss:
- i Z o t o p e ’ s M B I T + is even lower, buried, more like a roomtone:
- P S P ’ s X e n o n is like PT’s dither, but softer:
- W a v e s I D R T y p e 1 is very low level, almost nothing to hear.
- W a v e s I D R T y p e 2 is undetectable (at least for my ears on my Beyer Dynamic DT770 that is)
- B a r b a b a t c h : undetectable (see above)
So IDR and Barbabatch seem to be winners in terms of noticeability of added noise. Let’s also zoom in on differences between
IDR Types 1 and 2 in the L2 and L3 LL.
So now for the real deal, I generated some reverb tails in 32bit 44.1kHz, and repeated the process.
Indeed, here it is difficult to really hear any difference except, as before, IDR2 and Barbabatch leave no noticeable trace.
But when amplified with another 36dB, the L2 Type 2 Ultra looses out big time and seems to do weird stuff on different
channels:
And when zoomed in further, we see this:
The tail is being cut off with rough quantization. The L3 does a better job, but still you can hear tail and noise blend.
Barbabatch does look somewhat odd, but is definitely winner, hands down.

Conclusion:
I would not trust RX with samplerate conversion or even dither, although MBIT+ sounds pleasant and does the job, I’d go for
the L3 Type 2 Ultra. Pro Tools does a proper job at samplerate conversion (needles to say with Tweak Head) but its standard
dither is not all that good.
For dithering alone, Barbabatch seems to do a stellar job still, although it has been around for quite a while now.

Questions, reactions? Feel free to drop me a line…

Arno Peeters
[email protected]