Problem with Save Filter in AL 6.20

clarus

New Member
Thread Starter
Joined
Apr 21, 2021
Messages
15
When saving filter, AL would ask you to choose the sample rate to support. I usually select the 6 that I used: 44.1, 48, 88.2, 96, 176.4, 192. I generate it in both alc and zip.
My recent attempt seems to indicate that something is corrupted. When I playback at 176.4kHz/192kHz, something is out of phase. There is no more imaging and just diffused sound. Other sampling frequencies seems to work fine. I did noticed that when it was saving filter, the status bar at the bottom would normally show progress for every single sampling frequency. In this case, it will skip over 96kHz, and get stuck at 176.4kHz for like 20+ seconds before going onto 192kHz. It may have skip over the 96kHz because the measurement was taken at 96kHz and the filter has already been generated. The problem is applicable to both alc (when used with Convolver) or zip (when used with Roon).
I experienced this "out of phase" problem before but was able to work around it by resampling to a higher frequency, since the problem usually occur at 44.1 or 48kHz. But this time the problem happen at 176.4kHz (which all my DSD files are resampled to) and 192kHz.
I have tried both 32-bit float and 64-bit float filter format and the result is the same.
 
Last edited:

juicehifi

Audiolense
Staff member
Joined
Feb 5, 2018
Messages
575
The issue you are addressing is a general problem. When you save the filters for many rates in Audiolense, the filter length stays the same. This can become a limiting factor at very high sample rates, with unintended effects.

You can probably get around this by upsampling the measurement to 192 kHz before you create the corrections. Filter lengths and correction settings that work at 192 kHz will in most instances work equally well at lower rates. You may need a longer filter and possibly different settings for a good result at 176 and 192 kHz. When you see good result at the highest sample rate you should get good results for the others as well.

For Audiolense convolver I recommend that you only save for the sample rate that was used during measurement. The convolver will resample the filters, and their sonic behavior will be practically identical at all rates * except for the band with differences associated with each sample rate. If you e.g. make 65 k long filters at 48 kHz you will get 260 k long filters at 192 kHz. The convolver can handle such lengths with ease. Max filter length in the convolver is 65 k * 16. This is unlikely to be a limiting factor before the sample rate goes well above 1 million Hz.

It is a weakness in Audiolense that the generated filters don't have the excact same behavior at different rates, and that the behavior at all rates are not shown. You can, however open and examine the generated corrections. Most of the time you will see anomalies in the frequency response if something has gone wrong. Nevertheless I am contemplating whether to reample the correction filters in Audiolense, as it is done in the convolver. This will sometimes lead to arbitrary filter lengths and/or excessively long filters, and I am not sure how the various 3rd party convolvers will handle either.

Come to think of it; I will add such a variable-length resampling as an option and an alternative to the existing solution.
 
  • Like
Reactions: jjazdk and Mitchco

clarus

New Member
Thread Starter
Joined
Apr 21, 2021
Messages
15
Does it mean that ideally, I should do a measure at each sampling frequency and generate a filter for each measurement? For Roon zip file, I could combine the individual cfg and wav files. How about alc file?
 

juicehifi

Audiolense
Staff member
Joined
Feb 5, 2018
Messages
575
Cfg:

It means that you need to check that the correction procedure works for the highest sample rate. I doubt that there is anything to gain by measuring at all rates.

Alc:
create correction for one rate and let the convolver resample the filters.
 
  • Like
Reactions: whoareyou and Mitchco

whoareyou

Member
Joined
Dec 23, 2021
Messages
110
More  
Main Amp
Benchmark AHB2
Additional Amp
ATI AT525NC
Computer Audio
JRiver
DAC
Exasound E38
Front Speakers
Psb synchrony one
Center Channel Speaker
PSB synchrony one C
Surround Speakers
Revel m105
For me, I think this explains what I hear with resampled SACD playback at higher bitrate playback and AL filter (352.8 converted SACD via JRiver).

With AL filter at 352.8 it sounds as if I lose definition in the high end. Converting to lower sample rate, I eventually noticed when I resampled SACD to 48khz (same as measurement) everything sounded "right" to me, so that's rate I always use for listening. I couldn't tell difference between 48 and other slightly higher rates, so I settled for using that rate since it's a bit easier on the processor.

Now, after reading this I decided to see what my simulations look like at 352.8. I resampled my measurement to 352.8 and regenerated with current filter. Simulation now shows the frequency response with big drop at the higher frequencies (Guess only way I can validate is with a measurement, but I'm too lazy so I'll go by my ears with this one).

Other resampled higher rates, up to 192khz) mostly worked with the existing filter, but for 192 one small change to disable the "Prevent Treble Boost" option was required.

For me, simply increasing the filter length (over 192) did not fix the higher frequency dip.

If I am doing this correctly, this seems to explain what I am hearing. Otherwise.....
 
  • Like
Reactions: juicehifi
Top Bottom