Digital Acoustic Correction for Desktop & HiFi speaker systems using Room EQ Wizard and rePhase - P4
HTML-код
- Опубликовано: 18 янв 2025
- Part 4
This tutorial will take you through the process of generating a stereo impulse response filter based on measurements taken in your listening space. It is broken down into four main parts, which I hope will help make the segments easier to follow. The entire program time of the tutorial is slightly under one hour.
Your filter will be a 32-bit floating point WAV file, which can be loaded into just about every convolution engine around these days. Some examples include Roon, JRiver, HLC, SIR and SIR2, Convology XT, MConvolutionEZ, Impulser2, and more.
I just finished using this 4 part tutorial to tune my system. I've used some other REW methods (without RePhase) in the past, but these new filters are producing far-and-away the best sound my system has ever seen (or heard!). Thank you for taking the time to put this together.
fantastic tutorial. the sound in gorgeous. many thank's
You're welcome. Believe it or not it is possible to get even better sound out of these filters. I've been working on the process a lot since I posted these videos.
@@davidbrancato optimal. can you tell us how how? 😛
Thanks David : finally I can use REW & rephase correctly🙂.So far I was trying to flaten the phase in rephase without success. Adjusting the GD is physically making sense, but requesting a lot of effort : in my case the left loudspeaker in in a corner, largelly increasing low freq. Vector curves between right & left are very different in low freq (< 200hz). To ensure GDs are "perfectly" overlapping I had to play with the filter for 2 hours.... At the end, instrument are at the right place, local freq boost are amortized (saving a lot on RT60), and whatever is the freq it come from the same place for a given instrument. One more time thank you!
I guess your case is a bit more extreme than anything I have come across so far. I'm glad to hear that it worked out. Would you mind sharing some screenshots before and after? Also what does your step response look like before and after?
@@davidbrancato
Hello David,
I'm not familiar with youtube answers...how can I paste an image?
1)as a good start the left speakers : drive.google.com/file/d/1L7WEoTlxJc6UIo0IEYWNFMczTuQiWB6G/view?usp=share_link
one ca see the big boost in low freq : creating a kind of resonance, where we can't make differences between the frequencies. The room is a mix of rectangle and parallelograms at the extremities, 3.7m high (cathedral like), 250m3 volume : don't know how to determine modal freq...
2) original step : drive.google.com/file/d/1wNiFIgwZAz_f5Jw-Ulr9k52rtkS0BZ9_/view?usp=share_link
3)original GD : drive.google.com/file/d/1KL97i82LQvlnMJjY8lJWUiP3BeMf6frq/view?usp=share_link
4) predicted step after tuning : drive.google.com/file/d/1XgkZWrwWfHEFDS-FH9JI2R-e6Z1Py2aJ/view?usp=share_link
5) predicted GD after tuning : drive.google.com/file/d/1Qs_6FKkPAdJfQm-3Lrh9544LVR4T_-mh/view?usp=share_link
When I'll get time, I will redo new acquisition of freq sweep from 20 - 20K played with foobar & the pulse convolution...it should be "close" to the predicted, isn"t it?
Stephane
@@stephanephotos1767 I think you'll need to use dropbox or something other than Google Drive. It's denying access.
Are speaker distances accounted for in these filters, or would I need to add those in Roon speaker setup?
All the Vector processing - does that ever get loaded into RePhase? Did I miss domething?
Never mind, I saw the answer below
Awesome
Hi I also tried to linearize the phase with the paragraphic eq phase trying to bring everything closer to zero but comparing with the final result of your tutorial it turns out slightly worse even on the phase
As I recently explained to someone here, no attempt I made at modifying the phase during this process ever resulted in positive change, so I just don't bother. The filters always come out sounding better when I leave the phase how it is. To tell you the truth I don't even change the phase of the upper filter banks anymore. I just leave them all on minimum phase.
@@davidbrancato in fact, except for the very low frequencies which have 180 degrees of phase shift, the slope is very steep towards 0 and draws a straight line. absurd.
@@davidbrancato hi, only thing when I raise the volume a little more and only on some pieces the tweeter crackles which it doesn't do if I eliminate the convolution filtering. What could it be?
@@lucasantilli6510 I'm not sure. I'd have to see your mdat file if I'd have any chance of diagnosing it.
@@davidbrancato it seems that after a certain volume level the tweeter goes into saturation and starts to croak on some pieces quite loaded with bass
Wow! This is what I've been looking for.
Question: Is there anything to be gained by adjusting phase in Rephase? Thanks
I'm glad you found this helpful. I have been working on this process for a long time! Yes, there is a lot to gain from phase alignment, such as tighter, more refined bass and a more coherent soundstage. Does that answer your question or were you referring to something else?
How does it align for phase in rephase? It didn’t appear there were any adjustments made in rephase other than magnitude.
Curious, how this process is different than exporting a biquad file?
Why Blackman-Harris windowing?
Thank you David. Been trying to find out how to do this for ages. I adjusted my sample rates and taps to match the DSP I'm using. I got an error in rephase during impulse generation saying " Level was lowered by 0.96dB to avoid clipping in wav impulse format. Use IEEE-754 bin or float txt impulse formats if you want to avoid this issue". Do you know why this might be? I set my targets just below the lowest SPL in the frequency response.
Did you record the sweeps at your desired sample rate or did you record them at 48kHz and then switch it to something else when you generated the filters?
@@davidbrancato I recorded everything at 48kHz and maintained this through to the FIR filter. Although I just checked and my interface is set to 44.1kHz, but this would only be handling the output as I was using a USB UMM6 Mic, so I don't think that would have caused an issue...? The filters are working fine, so no dramas there.
@@mantolini It sounds like everything regarding the creation of the filters went fine, but I can't see what you did. I don't really know what could have caused that error you saw. I never got any errors myself. It sounds like one or more of your filter bands just poked up too high and rePhase turned the whole thing down automatically. If it only did this on one side, then your volume levels between the left and right probably wouldn't be balanced. Perhaps you could go through and take some screenshots of rePhase showing the two channels' settings/graphs and post them somewhere so I can look at them.
I'd be wary of the sample rate conversion happening if your output is still 44.1kHz. Are you able to change the output to 48kHz? Oh, and how does the filter sound?
@@davidbrancato The filter actually sounds fantastic. I did a post measurement and it's very flat from about 2kHz upwards. Unfortunately you can't filter out room modes so the mid bass response is a bit peaky, but it's not too bad at all and the sub bass is just where I want it. However, when I get some more time to play around I will take another pass at all the steps and send you a message with the results.
@@mantolini I know it does! I worked hard on this! But I'm curious... at what level did you set the target? Did you smooth the response to 1/6 or psychoacoustic and then place the target on the lowest dip? If you do that, you're more or less bringing the rest of the response down to match the level of the room modes. In that case, your whole spectrum should be within 5 or 6 dB of variance (+/- 2dB or +/- 3dB).
To prevent rePhase to import the filters backwards, just sort them ascending in REW before exporting them. ;)
Yes I figured that out! Thanks for bringing attention to it, though. As it turns out, I still prefer to put them in ascending order in rePhase anyway. This is because I usually re-organize some of the bands within the banks depending on what frequency they are, so in my case it doesn't matter.
@@davidbrancato Another question... are the filters generated by rePhase in your tutorial min-phase or linear-phase filters?
@@Cathul they are set up as mixed, with the low frequencies marked as min phase and the higher frequencies marked as linear. This is the reason for the multiple banks. Now when you export the combined filter in REW you have the option of selecting “measured” or “minimum phase”, which makes a difference as well. Make them both and see which you prefer.
Should I run my REW measurements at 96kHz resolution?
Also, how does this compare with parts 2 and 3 of this video? It’s similar but different. He generates and inversion filter and then convolves it with a phase adjusted export from Rephase. I don’t think there’s a method to put that inverted filter from Part 2 into rephase.
How is your method different than parts 2 and 3 of his in the end result?
ruclips.net/video/cgKeHyT7Xgo/видео.html
Also, your video is fantastic!!!
Run them at 48kHz. REW works internally at 48kHz and has a good resampling algorithm for exporting different resolutions.
You are correct. There is no way to bring an inverted response into Rephase. You have to bring the filter banks in like I do if you want to use my approach for inversion.
I don't really know how different my method is from his in terms of how the filters sound, but I can tell you that he is not dealing with the frequency response the same way I do it. He's only chopping the peaks off an unsmoothed response while I dig a bit deeper and level things out more with smoothing. I've never heard of processing the data unsmoothed as he does it. I find that strange, but I assume he gets good results, otherwise he wouldn't do it.
I tried FDW of a while, but settled on 1/6 and psychoacoustic smoothing instead because I like the results better. If you process the frequency domain at 1/6 or 1/12, and then deal with the group delay/step response with psychoacoustic smoothing applied to the result of the frequency inversion, it works incredibly well. No math or other complex operations going on, just match up the group delays and monitor the step response.