SRC - Weapon of choice?

[eneasheynen]

What's your preferred sample rate converter?
I used to use the one built in to the Pro Tools import window, but someone opened my eyes - it was pretty crap.

Since then I used the one in iZotope RX and sometimes SoX from the commandline, but I wondered whether or not there were better ones. I stumbled onto the infinitewave SRC comparison and nosed around a bit, but was pretty overwhelmed.

Now recently, Reaper introduced the r8brainfree algo into their batch converter (which i use for some stuff), and i looked it up on the infinitewave website - I was disappointed by what I saw (the filter was already cutting below 20kHz).

I then did my own test and noticed that the info on the website differed to my own (maybe an updated version or different settings?). My results were much better.
Now I wonder if it's worth switching for convenience... Also, the results were "technically better" than those from the iZotope SRC.

TL;DR should I switch my SRC from iZotope to r8brainfree built in to Reaper?
What do you use?
... Should I even care which one I use?

[mystran]

Now recently, Reaper introduced the r8brainfree algo into their batch converter (which i use for some stuff), and i looked it up on the infinitewave website - I was disappointed by what I saw (the filter was already cutting below 20kHz).

One can very roughly divide samplerate converters in the wild into "good" ones and "poor" ones. The "poor" ones are those that are significantly worse than what is possible (eg. poor passband shape, poor attenuation, or both), while the "good" ones are those that are based on solid engineering trade-offs, trying to choose a suitable balance between various goals. There are plenty of good SRCs and plenty of poor ones. It is interesting to note that the distinction between "poor" and "good" in this particular case tends to be very clear, the poor ones tend to be really poor.

Both iZotope RX and r8brainfree clearly belong to the "good" class and although the tradeoffs they choose are slightly different, they are all valid tradeoffs. Assuming that https://src.infinitewave.ca/ is at least roughly up to date (and even if it isn't I'd assume any updated versions won't be much worse), it seems that iZotope puts more emphasis on keeping a flat passband, while r8brainfree puts more emphasis on preventing aliasing completely at the cost of some high frequency attenuation. These are conflicting goals. SoX is both flatter and steeper than either of these, but due to the physics of how this works, it then has to pay for this with more latency and risk more preringing.

While the trade-offs made by all three are slightly different, they are all high quality algorithms created by people who clearly know what they are doing and ultimately the choice of which one you prefer is largely a matter of personal taste. There is no clear winner.

[Unaspected]

Been using R8Brain for a while - Love that it is now integrated in Reaper.

Cockos keep killing it.

[mystran]

I'd like to add that personally when writing SRC code, these days I'd probably go with a tradeoff most similar to r8brainfree, but it's important to understand that this is by no means "obviously better" and any of the converters mentioned are perfectly fine.

[kobal]

r8brainfree has not much ringing , i guess i won t need something else, it s great to have it in reaper..

[cantaloupe]

Thanks for dropping the knowledge, mystran. Good stuff.

While i rely on the recent r8brainfree addition in Reaper for media item resampling, i do however not trust Reaper rendering to downsample my 96k projects whatsoever. My bad experience was with the previous SRC algos however, where i experienced large differences between the sound in project vs export to lower sample rates.

I need to do more comparisons to see if r8brainfree has solved this issue, but as of now, i just render in 32bit FP wav from Reaper, then use SoX downsampling+dither+delivery_format as a last step. Automated via a simple terminal script, so it barely takes an extra second.

[Aleksey Vaneev]

Check out "r8brain 2" at https://src.infinitewave.ca/, not "r8brain free" which is antiquated, not even version is known, somebody submitted it without due diligence. I've contacted InfiniteWave to update the listing.

[mystran]

Check out "r8brain 2" at https://src.infinitewave.ca/, not "r8brain free" which is antiquated, not even version is known, somebody submitted it without due diligence. I've contacted InfiniteWave to update the listing.

I see.. well that's going to ring a lot more (although I guess not so much as to be a huge problem).

edit: Doesn't change the main point though, which is that these are all algorithms that are good enough (and I'm going to insist that holds even for your "antiquated" version) that someone listening to the results without knowing what was used is not going to be any wiser. Few people will have sufficiently accurate hearing in the 18kHz+ region anyway to audibly judge the differences and ultimately we're talking engineering tradeoffs here.

[jamcat]

Unfortunately, r8brain PRO is not on Mac and I guess never will be.
Good news is it looks like Acon Digital is once again killing it, and is just as good.

[Aleksey Vaneev]

That's not quite precise, r8brain PRO may be released for Mac sometime in the future. It's just true that with all accumulated advancements in SRC field, including fast r8brain free, r8brain PRO is a very niche product; releasing a Mac version won't change things commercially for us.