Old 11-15-2021, 06:28 PM   #1
Nixon
Human being with feelings
 
Nixon's Avatar
 
Join Date: Dec 2011
Posts: 406
Default v6.41+dev1115 - November 15 2021

v6.41+dev1115 - November 15 2021
  • * Includes feature branch: r8brain free samplerate conversion mode
  • * Includes feature branch: Windows very long pathname support
  • * Includes feature branch: media explorer crossfades
  • * Includes feature branch: recording files wildcard support
  • * Includes feature branch: video render/convert normalization
  • * Includes feature branch: media item notes improvements
  • * Includes feature branch: .caf media format support
  • * Includes feature branch: ASWG metadata support
  • * Includes feature branch: media item lanes
  • + ARA: update SDK to version 2.00, with ARM support
  • + Media explorer: display .RPP start tempo rather than project tempo setting [t=259656]
  • # Media item lanes: fix glue with multiple item lanes [p=2497445]
  • # Razor edits: fix extra copies [p=2497447]
  • # VST: fix recent FX matching for new ident strings [p=2497358]
  • # r8brain: fix incorrect output at end of multichannel rendered files
  • # r8brain: small tweaks to lookahead calculation
This thread is for pre-release features discussion. Use the Feature Requests forum for other requests.

Changelog - Pre-Releases

Generated by X-Raym's REAPER ChangeLog to BBCode
Nixon is offline   Reply With Quote
Old 11-15-2021, 07:52 PM   #2
Funkybot
Human being with feelings
 
Funkybot's Avatar
 
Join Date: Jul 2007
Location: New Joisey
Posts: 5,990
Default

Quote:
Originally Posted by Nixon View Post
[*]+ ARA: update SDK to version 2.00, with ARM support
Is there an ARM version of Melodyne out there to test this with? Or is that likely forthcoming and this is just the predecessor to that?
Funkybot is offline   Reply With Quote
Old 11-16-2021, 01:10 AM   #3
Dragonetti
Human being with feelings
 
Join Date: Feb 2017
Location: Kiel
Posts: 974
Default

Quote:
Originally Posted by Nixon View Post
v6.41+dev1115 - November 15 2021

[*]+ Media explorer: display .RPP start tempo rather than project tempo setting [t=259656]
Thanks , but
does not go right yet
the resolution is a bit too high

Last edited by Dragonetti; 11-16-2021 at 01:24 AM.
Dragonetti is offline   Reply With Quote
Old 11-16-2021, 01:50 AM   #4
Embass
Human being with feelings
 
Embass's Avatar
 
Join Date: Jan 2014
Posts: 923
Default

Embass is offline   Reply With Quote
Old 11-16-2021, 02:36 AM   #5
o_e
Human being with feelings
 
o_e's Avatar
 
Join Date: May 2016
Posts: 681
Default

o_e is offline   Reply With Quote
Old 11-16-2021, 02:37 AM   #6
Embass
Human being with feelings
 
Embass's Avatar
 
Join Date: Jan 2014
Posts: 923
Default

Embass is offline   Reply With Quote
Old 11-16-2021, 03:00 AM   #7
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

Quote:
Originally Posted by o_e View Post
Do you have your video decoder properties set up like this?



You need at least one of the first three (having only qt won't work).

I have (probably) the exact same file and here it works perfectly.



I'm on High Sierra though, not sure if something has changed on newer macOS versions.
Phazma is offline   Reply With Quote
Old 11-16-2021, 03:16 AM   #8
Embass
Human being with feelings
 
Embass's Avatar
 
Join Date: Jan 2014
Posts: 923
Default

Embass is offline   Reply With Quote
Old 11-16-2021, 03:29 AM   #9
o_e
Human being with feelings
 
o_e's Avatar
 
Join Date: May 2016
Posts: 681
Default

Quote:
Originally Posted by Phazma View Post
Do you have your video decoder properties set up like this?



You need at least one of the first three (having only qt won't work).

I have (probably) the exact same file and here it works perfectly.



I'm on High Sierra though, not sure if something has changed on newer macOS versions.
That did the trick, thanks a lot!
o_e is offline   Reply With Quote
Old 11-16-2021, 03:50 AM   #10
Embass
Human being with feelings
 
Embass's Avatar
 
Join Date: Jan 2014
Posts: 923
Default

Embass is offline   Reply With Quote
Old 11-16-2021, 03:56 AM   #11
EvilDragon
Human being with feelings
 
EvilDragon's Avatar
 
Join Date: Jun 2009
Location: Croatia
Posts: 24,790
Default

Quote:
Originally Posted by Nixon View Post
VST: fix recent FX matching for new ident strings [p=2497358]
Aha! So I was on to something after all Thanks, devs!
EvilDragon is online now   Reply With Quote
Old 11-16-2021, 07:41 AM   #12
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

Quote:
Originally Posted by Nixon View Post
# r8brain: fix incorrect output at end of multichannel rendered files
Ok, that fixed it!

Quote:
Originally Posted by Nixon View Post
* Includes feature branch: r8brain free samplerate conversion mode
I have done some general testing and comparison of all SRC modes, converting a pretty generic pop-house production from 44.1KHz to 48KHz. The music has a less dense verse part with rather little limiter action and a more dense chorus part with a bit more (true peak) limiting. All that processing has of course been done during production and the SRC in these tests has acted directly on the already rendered 44.1KHz master file.

In terms of render speed I got following results:

1) Lowest: 121.7x
2) Low: 118.3x
3) Fast (IIR + Lin): 110.3x
4) Fast (IIRx2 + Lin): 98.3x
5) Fast (Sinc): 88.0x
6) r8brain: 57.8x
7) Medium: 57.8x
8) Good: 29.1x
9) Better: 25.6x
10) HQ: 20.9x
11) Extreme HQ: 15.5x

In terms of null test peaks against Extreme HQ I got following results (1st place = lowest peak, meaning most cancellation and thus closest result to Extreme HQ):

1) Extreme HQ (obviously, 100% cancellation)
2) HQ
3) Better
4) Good
5) Medium
6) r8brain
7) Fast (Sinc)
8) Low
9) Fast (IIR + Lin)
10) Fast (IIRx2 + Lin)
11) Lowest

In both tests r8brain performs almost identical to Medium (64pt Sinc). In the the peak cancellation test the peak difference is about 1 to 3dB.

Doing the same peak cancellation test but with r8brain against all others I get almost identical cancellation against all modes from Extreme to Medium. Those that performed worse than Medium in the original test, also performed worse against r8brain in this test, in the same order.

So according to all my tests, r8brain seems to deliver pretty much the same results as Medium (64pt Sinc). I am wondering, in practical application like converting a piece of music, is there a way to show that r8brain is superior to Medium (or all Sinc modes for that matter), not only in regards to speed?

Some other things I noticed about all these SRC modes in general (with the material I fed them):

From a listening perspective all the Sinc modes and r8rbain sound almost the same.
The Linear Interpolation modes cause a noticeable loss of high frequencies.
The Point Sampling mode adds some harsh high frequencies.
I'd argue that for dense material containing much high frequencies (like EDM or Rock/Metal with distorted guitars) the Point Sampling mode is superior to the Linear Interpolation modes as the added high frequencies don't stand out that much.
For less dense material with less high frequencies (like Acoustic, Classical, Vocal or maybe also more laidback Pop, House and Rap songs) the Linear Interpolation modes are better than Point Sampling as the high frequency loss is less noticeable than the harsh added high frequencies would be.
I'd agree with Justin's answer on the previous pre-release thread though that all non-Sinc modes (except r8brain) are best avoided for serious conversion duties.

Another thing that I want to point out are max peak (and visible waveform) changes with the various modes, which might lead to clipping if someone is unaware.
If using a True Peak limiter on the original render, r8brain and all Sinc modes (except fast) seem to end up on the same max peak value. If the original render however has true peak clips, the converted audio file may end up with higher and thus clipped actual sample peaks. My guess is that the Sinc SRC algorithms perform some oversampling to construct missing samples and oversampling leads true peaks to become actual, existing sample peaks.

Interestingly, all the Fast modes are affected by the converted file having lower peaks by several 0.xdB than the original file, which might even lead to slightly lower LUFS levels. If the conversion is done on a file which has not been true peak limited, they can have higher peaks instead.

The Point Sampling mode seems to maintain the original sample peak values no matter if true peak clipping has occured or not. Probably it doesn't do any oversampling. However I wouldn't recommend Point Sampling for most circumstances for above mentioned reasons.

A lot of this post probably has to do little with the current pre-release but I hope that these tests and my interpretations make sense and might help others to get an idea about the various SRC modes in Reaper. Also, maybe it helps the devs to check if everything performs as intended. If there are flaws in my tests and interpretations please let me know.
Phazma is offline   Reply With Quote
Old 11-16-2021, 09:41 AM   #13
Aleksey Vaneev
Human being with feelings
 
Join Date: May 2010
Posts: 28
Default

Quote:
Originally Posted by Phazma View Post
So according to all my tests, r8brain seems to deliver pretty much the same results as Medium (64pt Sinc). I am wondering, in practical application like converting a piece of music, is there a way to show that r8brain is superior to Medium (or all Sinc modes for that matter), not only in regards to speed?
Sorry, but that's a bold statement. Peak difference has nothing to do with audible quality, it mainly reflects how filters are constructed.

I suggest you to create 96000 project, load some 44100 file, put SPAN at master bus with display range up to 48kHz (and HiRes Window, slope 9 for better view), and compare different "playback resampling" modes, you'll see how they differ not only in peak difference.

Last edited by Aleksey Vaneev; 11-16-2021 at 09:53 AM.
Aleksey Vaneev is offline   Reply With Quote
Old 11-16-2021, 11:03 AM   #14
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

Quote:
Originally Posted by Aleksey Vaneev View Post
Sorry, but that's a bold statement. Peak difference has nothing to do with audible quality, it mainly reflects how filters are constructed.

I suggest you to create 96000 project, load some 44100 file, put SPAN at master bus with display range up to 48kHz (and HiRes Window, slope 9 for better view), and compare different "playback resampling" modes, you'll see how they differ not only in peak difference.
I didn't intend to claim that r8brain isn't superior to the other Sinc modes. It is just that with my audio material and my way of testing I couldn't find out in what way it is superior and I am curious to know so that I can identify and appreciate its benefits. Actually I am curious in general how to test and what to listen for when comparing SRC algorithms, so that I can consciously choose the right one for my purposes. I find it hard to tell by ear but maybe I just don't have the right audio material.

So sorry it came over as a statement (or even a judgement of your work), it was actually meant as a question, coming from me not knowing much about SRC and being curious to learn more.

I have tried the test you suggested. The difference that I could spot in the various SRC modes is the frequency content above 22.5KHz (so the Nyquist frequency of the original 44.1KHz file samplerate). The higher quality the Sinc mode, the lower the energy of the frequencies in this area and r8brain got the lowest energy, thus I guess best result.

So my (bold) guess is that SRC in general introduces aliasing and the better the algorithm the less of it is there? To test this I have rendered the 44.1KHz file to 96KHz with both r8brain and Medium 64pt Sinc and then converted both these renders back to 44.1KHz with the same algorithms and did the polarity switch test against the original, non-converted 44.1KHz file.

As expected, SPAN shows much more activity (=spectral difference to the original) in the audible range in the file converted with the Medium 64pt Sinc mode than it does in the file converted with r8brain. So I guess this means r8brain outperforms the other algorithms in terms of aliasing?

MEDIUM 64pt sinc:



R8BRAIN:



I hope this test made more sense and did r8brain justice. It is still hard for me to pick out the differences by ear but I think I have understood that what I should be listening for are harsh/inharmonic frequencies in the top end. Perhaps if I try with acoustic music I might detect it more easily.
Phazma is offline   Reply With Quote
Old 11-16-2021, 11:34 AM   #15
Aleksey Vaneev
Human being with feelings
 
Join Date: May 2010
Posts: 28
Default

Quote:
Originally Posted by Phazma View Post
I hope this test made more sense and did r8brain justice. It is still hard for me to pick out the differences by ear but I think I have understood that what I should be listening for are harsh/inharmonic frequencies in the top end. Perhaps if I try with acoustic music I might detect it more easily.
Thanks for such test, yes it's more revealing. It's not only about aliasing, but as your SPAN images show, it's also about broadband precision.

I can clearly hear a difference between Sinc and r8brain in 96000 project with 44100 mediafile. On the other hand, I bet some users will still prefer Sinc - "coloration" of sample rate converters is a debatable topic, "harshness" may be preferable in some genres.
Aleksey Vaneev is offline   Reply With Quote
Old 11-16-2021, 12:57 PM   #16
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

Quote:
Originally Posted by Aleksey Vaneev View Post
Thanks for such test, yes it's more revealing. It's not only about aliasing, but as your SPAN images show, it's also about broadband precision.

I can clearly hear a difference between Sinc and r8brain in 96000 project with 44100 mediafile. On the other hand, I bet some users will still prefer Sinc - "coloration" of sample rate converters is a debatable topic, "harshness" may be preferable in some genres.
Thanks for the explanation! It looks like the r8brain frequencies decrease in a more linear way, introducing thus less coloration than the less precies Sinc modes.

Interesting what you say about different opinions about SRC coloration and it being wanted in certain circumstances. I can see added harshness being a bonus in genres that use bright synths and/or distorted guitars as the harshness may be interpreted as added presence, whereas in genres based on acoustic instruments and voice it may lead to an unnatural sounding record and higher quality SRC is a benefit.

I will try to listen for this with some more different types of audio material and see if I can hear the coloration and learn to recognize it. Different SRC may be good for different purposes and I find it awesome that Reaper gives us so much choice and especially that you have given Cockos the possibility to integrate r8brain.

According to spectral analysis r8brain delivers even slightly more precise conversion than the 768pt HQ Sinc mode, while being much more CPU efficient (rendering about as fast as the 64pt Sinc mode). I think it would be the best default choice for most users and probably should become Reaper's new default mode.

Thanks again to Aleksey Vaneev and Cockos for bringing r8brain to Reaper!
Phazma is offline   Reply With Quote
Old 11-16-2021, 01:12 PM   #17
BPBaker
Human being with feelings
 
BPBaker's Avatar
 
Join Date: Oct 2013
Location: Brooklyn, NY
Posts: 209
Default

I'm getting a reproducible crash in dev1115 on MacOS 10.14.6 whenever attempting to make a razor edit in a new project without tracks. (No crash in 642rc1 doing the same thing, though.)
BPBaker is offline   Reply With Quote
Old 11-16-2021, 03:57 PM   #18
Justin
Administrator
 
Justin's Avatar
 
Join Date: Jan 2005
Location: NYC
Posts: 15,721
Default

Ah heh you didn't have to delete your reply to mine Phazma, I just had continued reading the thread and figured my response was redundant

It will be a while before we change the default modes, we'll have to get this all extremely well-tested and consider the other implications. But I think the next build should have the combo boxes look about like this:
Justin is offline   Reply With Quote
Old 11-16-2021, 04:29 PM   #19
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

Quote:
Originally Posted by Justin View Post
But I think the next build should have the combo boxes look about like this:
This already looks much clearer than it is currently! I’d say this could even be released before r8brain if the r8brain implementation needs much more testing. But perhaps it makes more sense to release as many SRC related features as possible in a single update.

While your screenshot definitely goes in the right direction, in my opinion it could still be tweaked a bit to be even clearer and more informative. Some modes have comments about quality, others about speed, r8brain and Point Sampling have a useful comment about application and several modes have no comment at all.

Personally I would probably put information about quality, speed and perhaps a brief comment about suitability/usecase in brackets after all modes and do so in a consistent order (eg. quality indication always first). That would make it very easy for the user to understand how they are sorted and which one best to pick and it would also look a bit more tidy.
Phazma is offline   Reply With Quote
Old 11-16-2021, 04:48 PM   #20
Funkybot
Human being with feelings
 
Funkybot's Avatar
 
Join Date: Jul 2007
Location: New Joisey
Posts: 5,990
Default

I read Justin's screenshot trying to communicate "worst to best." But I'm not 100% sure that's intent. Is "Linear Interpolation + IIR x2" worse than the lowest quality Sinc setting? If it really is a worst to best list, then great. If not, it comes across that way and maybe more info elsewhere is a good idea.

I do love how the R8Brain message specifically calls out what it's good at and where you don't want to use it. Same for Point Sampling. I find those types of lay-person descriptions super helpful.
Funkybot is offline   Reply With Quote
Old 11-16-2021, 05:20 PM   #21
Justin
Administrator
 
Justin's Avatar
 
Join Date: Jan 2005
Location: NYC
Posts: 15,721
Default

Quote:
Originally Posted by Funkybot View Post
I read Justin's screenshot trying to communicate "worst to best." But I'm not 100% sure that's intent. Is "Linear Interpolation + IIR x2" worse than the lowest quality Sinc setting? If it really is a worst to best list, then great. If not, it comes across that way and maybe more info elsewhere is a good idea.

I do love how the R8Brain message specifically calls out what it's good at and where you don't want to use it. Same for Point Sampling. I find those types of lay-person descriptions super helpful.
Yeah, 16pt Sinc is preferable to the IIRx2 mode IMO. The ordering is low-to-high quality.

Quote:
Personally I would probably put information about quality, speed and perhaps a brief comment about suitability/usecase in brackets after all modes
It's way too busy if they all have comments, and what can one say about the differences between 192pt and 384pt sinc? "slightly slower, slightly better" ?
Justin is offline   Reply With Quote
Old 11-16-2021, 05:56 PM   #22
TonE
Human being with feelings
 
Join Date: Feb 2009
Location: Reaper HAS send control via midi !!!
Posts: 4,031
Default

Quote:
Originally Posted by Phazma View Post

While looking at this gif what I thought was: Would not it be cool if you could do same with midi files as well? On the right a list of midi files, and you drag it via some hotkey, e.g. Control-drag to arrange, and Reaper will auto-render it using current track as audio, not midi. If you want midi just do normal drag (without additional Control).

Maybe interesting for the devs, increasing the coolness factor of Reaper (as if Reaper is not supercool yet.
TonE is offline   Reply With Quote
Old 11-17-2021, 02:26 AM   #23
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

Quote:
Originally Posted by Justin View Post
It's way too busy if they all have comments, and what can one say about the differences between 192pt and 384pt sinc? "slightly slower, slightly better" ?
Personally I would do it somewhat like this (not sure if it is technically all correct, it's just to give an example):

Point Sampling........................(lowest quality, fastest) - retro
Linear Interpolation................(very low quality, very fast)
Linear Interpolation + IIR........(very low quality, faster)
Linear Interpolation + IIRx2....(very low quality, fast)
Sinc Interpolation: 16pt..........(low quality, rather fast)
Sinc Interpolation: 64pt..........(medium quality, medium speed)
Sinc Interpolation: 192pt........(high quality, slow)
Sinc Interpolation: 384pt........(high quality, slower)
Sinc Interpolation: 512pt........(very high quality, very slow)
Sinc Interpolation: 768pt........(very high quality, slowest)
r8brain free..............................(highest quality, medium speed) - unsuitable for dynamic varispeed

Left out on the application/characteristics comments except the ones you already made as I didn't know what would be most appropriate. Based on my experience "HF Loss" would describe the Lin Int modes pretty well.

I think this way it would be pretty clear that they are sorted based on quality and also where r8brain lies in terms of speed, despite being the highest quality algorithm. Also some space between name and description (I placed dots because the forum doesn't let me do otherwise) would probably help making it look less busy and more organized.

Of course it's just a raw idea, up to you Justin to decide if something like this makes sense to you or you prefer to keep it as in your proposal.
Phazma is offline   Reply With Quote
Old 11-17-2021, 05:48 AM   #24
Aleksey Vaneev
Human being with feelings
 
Join Date: May 2010
Posts: 28
Default

Quote:
Originally Posted by Phazma View Post
Of course it's just a raw idea, up to you Justin to decide if something like this makes sense to you or you prefer to keep it as in your proposal.
Looks reasonable. r8brain on Apple M1 may actually be "rather fast", if Reaper enabled "pffft double" FFT algorithm.
Aleksey Vaneev is offline   Reply With Quote
Old 11-17-2021, 06:14 AM   #25
Phazma
Human being with feelings
 
Join Date: Jun 2019
Posts: 2,872
Default

If it manages to be faster than even the simplest Sinc mode that would be pretty cool
In my example I placed it at medium speed as it was as fast as Medium Sinc in my test but that may of course depend on several factors so not sure how it looks on paper and how the speed of the various modes can be compared objectively, if that’s even possible (I’m on Mac Intel btw).
Phazma is offline   Reply With Quote
Old 11-18-2021, 01:40 AM   #26
Tale
Human being with feelings
 
Tale's Avatar
 
Join Date: Jul 2008
Location: The Netherlands
Posts: 3,646
Default

Quote:
Originally Posted by Aleksey Vaneev View Post
r8brain on Apple M1 may actually be "rather fast", if Reaper enabled "pffft double" FFT algorithm.
Yeah, but I think that would add a few hundred kB to the binary. Also, if it was me, then I'm not sure I would want to maintain the extra, optional code paths for SSE etc. But I guess it could still be worth it, depending on the speed-up.
Tale is online now   Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -7. The time now is 01:23 AM.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.