Spectacular work Alec! That is incredible problem solving and deduction.
On a personal note, you have proven it is not all in my head…
2 Likes
You’re not alone 
Thank you very much for your efforts and deduction skills Alex. And to the others that join in as well too, all appreciated. Look forward to a better sounding Roon in due course! 
1 Like
You’re certainly right: i’ve just ran the test in HQPlayer after switching on dithering (NS1). The correlation depth is similar to Roon around 150dB (for a 24bit file, 24x6=144dB dynamic range).
Quite interestingly, the spectrum of the “dithers” are exactly the same! Temporally they look a bit different though, Roon dither being a “1 bit” oscillation with an offset of 4, HQPlayer dither being centered on 0 and varying beetween -4 and 4. Certainly different algorithms. Would be curious to know which dithering method is applied by Roon internally at the time of truncation of the 64bit internal format.
In conclusion, Roon and HQPlayer behave according to logic. This leaves us with the only issue of improper filter resampling by Roon.
Amazing topic, and amazing work being done here !
Especially at you, @alec_eiffel !
One thing : you have discovered, what we ‘should’ already have known.
Please see @brian 's first post in this topic :
…people had (or should have had) made sure to play only 44.1kHz audio, through the provided 44.1kHz filter.
I know, that I did. I can read, that at least @ogs, and @Leong_Kin_Foong1 did take care of that too. But still claim audibility.
Further on : It seems that you are still measuring a slight degratation, when processing audio though a NULL Roon filter, in a corresponding sample rate.
This deviation is larger, than with processing through HQplayer. Did I understand that right ?
Is it possible to make any statement on its audibility, already ?
I’m very sorry, I cannot be available for further analysis today… 
.
Very well spotted @Marco_de_Jonge. However it is simply brilliant work by @alec_eiffel to have the re-sampling uncertainty confirmed by measurement.
I am looking forward to listening to the samples. I have not done this kind of AB in a long time. Must admit that I find the logic around testing methodology difficult so I’m glad others know how to do this!
See my last post just above: the difference comes from the dithering, and is absolutely normal. So Roon is not underperforming HQPLayer when given a Null IR at the song’s sampling rate, according to my measurements. HQPlayer is more configurable and offers a choice of dithering algorithms, whereas Roon is a black box. Not a big deal I think. People fond of tinkering will buy HQPlayer.
On my office system (2 Divatech Diva 16 bookshelves and ES9018 Audiophonics DAC): can’t hear a difference.
That’s what I did, and I heard differences, that’s why I did not consider an IR resampling issue : my file was 44.1k and the filter was 44.1k. Until I realised yesterday that the 44.1k/24bits TIDAL tracks I was listening were actually MQA, and upsampled to a higher rate! Forgot to check the signal path…
3 Likes
The difference I heard was with 24/44.1 material. This is sent as 24bit out of Roon both without and with DSP processing (according to Signal Path in Roon). When I tested I found that it is not necessarily the convolution that makes the difference, but something else within DSP. I could hear a difference with DSP (peq) enabled and convolution disabled. Maybe there is something common with @alec_eiffel’s latest findings here?
Clear, helpful. I really admire your level of honesty and open-ness, @alec_eiffel !
Summarizing :
- You (person) have not heard the difference, when processing through a filter with corresponding sample rate;
- You (person) have not found an indication for audible trouble in the same case, with measurement or analysis tools.
- You DID find an imperfection in the filter resampling algorithms, which Roon team should at least be aware of.
Then : the original question of the topic starter, is still completely open. No answer whatsover, yet :).
Just your (and my) anecdotal auditory experiences, coupled with measurements that show no evident trouble.
Same for the experience of @ogs, and perhaps other contributors (I cannot know for certain, the sample rates of their audio files).
Let’s just ask something, to be completely sure of the topic of investigation. Starting with @Leong_Kin_Foong1 (others can answer as well):
- You have some audio files, which allow you to hear this difference.
Can you confirm :
- these are all “real” 44.1kHz files ?
- They are all NOT MQA files ?
- You are NOT resampling yourself, in any way ?
@ogs : I do not see commonality between your latest mention, and Alec’s findings. You are not resampling, here.
The 24/44.1 track I use is delivered by Qobuz as such. It is not MQA and there is no resampling involved in Roon.
I know there is no direct connection between Alec’s resampling findings an my observations, but I suggest that if there is some error within DSP, this might also affect other parts of the processing.
I was primarily using a 48K/24bit AAC Tidal file for my test.
Its the Jess Glynne “My Love (Acoustic)” I mentioned earlier. The Tidal link is https://tidal.com/album/53465936
My convolution zip file contains 48K filters along with 44.1, 88.2, 96, 176.4, 192, 352.8, 384 and 705.6 filters so in my case there should have been no resampling.
(Back again from children’s birthday parties… 
)
Great, @Leong_Kin_Foong1 ! Theoretically, that means that Alec’s findings should NOT apply in your situation.
Simpler : Both Alec and myself, do NOT expect to hear an audible difference in the case that you describe.
Neither is anything relevant expected in measurements.
(Not wanting to speak for Alec : I’m sure he’ll correct me when necessary).
So : if you want to know if your observations are “real”, I’m afraid that you’ll still need a statistically validated A/B test of some sort… :(.
Some readers seem to read @alec_eiffel 's conclusions, as ‘there IS a sonic difference’.
I think it MUST be stressed : this is NOT his conclusion !!
It is only the case, when samplerate of your audio is different from the samplerate of your filter, AND if you ‘forget’ to include your filter in all possible sample rates.
(Both conditions must apply simultaneously, for ‘bad behaviour’).
There is still one thing that may apply in @Leong_Kin_Foong1 's situation.
I mean, something that Roon may be doing wrong.
Could you (Leong) please upload your entire NULL filter ? I mean : the zip set, containing all sample rates ?
The reason for asking : it MIGHT be, that Roon is accidentally picking the wrong filter (in the wrong sample rate).
I have not tested that, and cannot rule it out.
Both Alec and I WOULD expect at least measurable effects from that. I agre, that it could potentially even be_audible_.
I think @alec_eiffel is better qualified for that analysis, than I. I would kindly like to ask for your assistance, please ?
If nothing comes out of that : I see no more other options, than proper A/B testing.
By the way : HUGE compliments to all contributors, moderators, Roon, and the community in general.
On most other forums, this would have ended in a big flame war, in 30 minutes.
Here, we are in it for almost 33 full days. Hardly a single bad word. Everyone is reasoning, seeking solutions, and being respectful.
It should not be special, but it really is !
At this rate it sounds like the whole thing will be sorted before we need to do any listening tests🤪
Which would be a shame, I was quite looking forward to it!
1 Like
Here is the zip file Roon Null filters for all frequencies:
I wonder if it’s as simple as Roon is resampling even though it doesn’t have to and then encountering the AE bug?
Love this version of this track.
Yes, she’s got such an emotional voice and the arrangement is pretty simple so I figured it’s easier to hear differences.
Oh, and she’s a good artist too! 
1 Like
@Leong_Kin_Foong1 these are not NULL files but room correction files, mostly in the bass region as far as I can see in REW.
Here is a 44.1K NULL convolution file Dropbox - NullIR44-1K.wav - Simplify your life
You can upload it in Roon convolution engine and tell us if you hear a difference toggling convolution on and off.
Ah, did I make a mistake linking the wrong set of filters? Sorry.
Will download your one and try tonight. I take it I have to stick with a 44.1 file.
Confirmed with Thierry @ HAF that the zip set he attached to me was indeed a working convolution filter set (one of many versions) and I mistakenly thought it was a NULL set when he only send me a single 44.1 NULL filter. If I got it right this time, then I don’t have NULL filers for all frequencies but only one 44.1 version.
I believe my original 48K/24 bit NULL test was done with the single 44.1 NULL filter. If so, then I would have listen to the AE bug and that might account for the difference I heard. Will check again tonight with both a 48K and 44.1 file.
1 Like
If you did play a 48kHz file through the 44.1 filter Roon would have to resample. I am sure this has a bigger impact on sound than DSP alone. Will be interesting to know what you find.