REW Beta Release REW API beta releases

[John Mulcahy]

Actually, should these dips around 415 and 600hz really not be considered notches?

The inversion was generated with 30 dB of gain allowed, that's very extreme. Was it intentional or a typo?

 

[Lowclock]

The inversion was generated with 30 dB of gain allowed, that's very extreme. Was it intentional or a typo?

I was just kind of testing it out and didn't consider that it would matter since it didn't end up with nearly that much gain anyways. I see now that it apparently does make a difference to when exclude notches actually "kicks in". The way the gain limit just chops the top of the response straight off compared to the kind of compression effect you get with the old regulation control doesn't "feel" right, but I don't think you would've changed it like this if it was just worse. I probably just need to mess with it some more.

I definitely like that it works with smoothed responses now since otherwise the exporting and importing you had to do to accomplish that was a pain and had some weird effects sometimes. E: I don't quite know what to think about the way this interacts with exclude notches now, though. It seems kind of counterintuitive at first glance to invert a smoothed response and see exclude notches chop up the response to account for notches that practically don't even exist anymore. Maybe I'm just thinking about it the wrong way and the one that looks worse actually measures/sounds better. I need to find some time to pull out the mic and stuff and actually see how they look when measured, even though it's probably negligible.

 

[John Mulcahy]

I have modified it to deal with the high gain situation, that will be in the next build.

It seems kind of counterintuitive at first glance to invert a smoothed response and see exclude notches chop up the response to account for notches that practically don't even exist anymore.

Your smoothed response isn't the room's response, the notches are still there in the room's response and shouldn't be boosted.

 

[Lowclock]

Your smoothed response isn't the room's response, the notches are still there in the room's response and shouldn't be boosted.

Yes, definitely, but maybe I just misunderstood the concept to begin with. I thought you just shouldn't try to equalize/invert notches to match the rest of the response because they don't respond linearly or at all and some can vary a lot depending on position, not that you should so specifically carve them out of much broader tonal changes higher in the response so they aren't boosted at all. I guess it is kind of exaggerated in my example because it's in a car with only a few measurements and some poor technique. In some other testing like in a real room with more and better measurements it does seem to behave much more like I would expect and doesn't look like a zipper past 1k, haha. Garbage in, garbage out I guess.

 

[serko70]

After lots of debugging, I figured the "last" one of the frequency magnitude values is missing from impulses exported as text vs fetched and decoded magnitude response.

 

[John Mulcahy]

Builds updated today (beta 20) with these changes:

• Changed: Improved exclude notches behaviour when inverting with high maximum gain
• Changed: IMD and TD+N calculations tolerate clock rate differences up to 200 ppm
• Changed: Show clock rate difference for sine tones when not FFT locked using detected peak frequency
• Changed: Show a warning if "Get fundamental from generator" is being used but there is a clock rate difference

 

[John Mulcahy]

After lots of debugging, I figured the "last" one of the frequency magnitude values is missing from impulses exported as text

I guess you mean measurement exported as text rather than impulse. Do you have an example mdat showing that?

 

[serko70]

I guess you mean measurement exported as text rather than impulse. Do you have an example mdat showing that?

Yes, sorry measurement not impulse in the mdat exports (with default options) with last freq mag missing compared to the decoded values with this algorithm:

const magBase64 = await fetchSafe('frequency-response',i);
const bytes = Uint8Array.from(atob(magBase64.magnitude), c => c.charCodeAt(0));
const buffer = bytes.buffer;
const floatArray = new Float32Array(buffer);
const dataView = new DataView(buffer);
const arr = [];
for (let j = 0; j < floatArray.length; j++) {
const splValue = dataView.getFloat32(j * 4);
arr.push(splValue);
}


It's 1/1 oct smoothed in the mdat but other smoothing parameters behave the same.

 

[John Mulcahy]

Thanks. A precision problem which caused the last frequency to be just past the end of the data, I've fixed it for the next build.

 

[achugh]

@John Mulcahy I was using the v5.40 Beta 19 build and noticed the following issues:

1. In the EQ panel, under "Filter Tasks" once I click on "Match response to target" and REW generates the EQ filters (in my case miniDSP 2x2 HD); if I try to edit any of these filters manually by opening the EQ Filters panel and then click on "Measure with these filters". The measured response that is saved as a new measurement adds an EXTRA "Hz" in the NOTES area of the measurement of every filter I touch (maybe all of them even though only 1 filter was touched). I paid attention to it afterwards.
2. I was having issues with REW crashing. I connected to my AVR via HDMI and selected the JAVA EXCL drivers to send the sweep signal. While connected to AVR, I also connected to miniDSP via USB which also shows up as EXCL driver in REW. In this situation, I went into preferences and changed the output device from AVR to miniDSP but forgot to change the input from ANALOG to USB in miniDSP interface. When I clicked on CHECK LEVELS on the measurement tab, REW froze as it heard nothing since the sound was going to miniDSP on the USB input but miniDSP was configured for ANALOG input. Trying to take a measurement also froze. Once REW crashed and I restarted, luckily it picked up my measurements from its TEMP folder on the WINDOWS machine. However, at this point every time I try to take a measurement or check levels, it continued to freeze. The only way I was able to make REW work again was to open REW, then close it without taking any measurements where it showed me the window "Saving Preferences" and then starting again, made it work. In other words, starting up REW after a crash made it unstable.

Can I request a new feature? Is it possible to tell REW to HOLD on to certain PK values and try the rest instead of always switching the complete set when I click on "Optimise gains, Qs and Frequencies"? As an example, I liked some areas of the EQ but others I wanted REW to give it another shot at finding a different combination.

 

[John Mulcahy]

The measured response that is saved as a new measurement adds an EXTRA "Hz" in the NOTES area

Thanks, I have fixed that for the next build.

I was having issues with REW crashing.

Please use the "Generate diagnostic file" option in the Help menu and attach the file.

Is it possible to tell REW to HOLD on to certain PK values

Deselect the "Auto EQ" setting for any filters you want left untouched.

 

[HD MOVIE SOURCE]

After lots of debugging, I figured the "last" one of the frequency magnitude values is missing from impulses exported as text vs fetched and decoded magnitude response.

View attachment 69153

I see this issue as well. I typically manually fix it.

 

[HD MOVIE SOURCE]

John, any chance we can have the option to export measurements as text without phase information included?

 

[jtalden]

Error trying to save any captured graph image:

 

[achugh]

@John Mulcahy I was taking subwoofer measurements when, this measurement window got stuck and did not show any sweep preview in the bottom left even though I heard the whole sweep and the reference chirps. I am attaching the diagnostic file for your re

 

[John Mulcahy]

I was taking subwoofer measurements when, this measurement window got stuck and did not show any sweep preview in the bottom left even though I heard the whole sweep and the reference chirps

Looks like a mic problem.

 

[John Mulcahy]

Error trying to save any captured graph image:

Thanks, I've fixed it for the next build. A workaround in the meantime is to set the comment position to be above or below the graph rather than on it. It's handy if you can use the "Copy to clipboard" button and attach the text when reporting an error.

 

[HD MOVIE SOURCE]

I sent an email to Mini-DSP about the SHD because I'm thinking about buying one for the XLR outs. I wondered if it has higher Q resolution for EQ vs the Mini-DSP 2x4HD, this is what they said.


"We run SHD DSP @ 450MHz for the Dirac 96kHz resolution . Sonic Quality of HiFi gears not only depends on the Key components. SHD have much better voltage regulation circuitry ( some stages cascaded) and the highly complex ADC and DAC stages in best layout arrangement.
SHD is our flagship model in our product ranges.

Our Answer is : SHD has better resolution than the 2x4HD.
https://www.minidsp.com/products/streaming-hd-series/shd please read SHD product page for more details."


With this understanding, how would I get improved EQ performance in REW when using an SHD vs the 2x4HD? For example, instead of being locked to 50.0 Q resolution, can the SHD use higher Q resolution? Like 60.0 Q for example? I'm obviously only asking because I have no idea how this increased performance effects EQ filters. However, the SHD is still 96kHz, so is this the Q limit for 96kHz?

Thanks.

 

[John Mulcahy]

John, any chance we can have the option to export measurements as text without phase information included?

I have added an option for that in the next build.

 

[John Mulcahy]

I sent an email to Mini-DSP about the SHD because I'm thinking about buying one for the XLR outs. I wondered if it has higher Q resolution for EQ vs the Mini-DSP 2x4HD

I think you confused them by referring to "Q resolution" when it seems you are interested in the maximum Q (and hence the narrowest filter). A maximum Q of 50 is already very high, however, I can't think of circumstances that could require such high Q values.

 

[HD MOVIE SOURCE]

I think you confused them by referring to "Q resolution" when it seems you are interested in the maximum Q (and hence the narrowest filter). A maximum Q of 50 is already very high, however, I can't think of circumstances that could require such high Q values.

Good to know.

 

[HD MOVIE SOURCE]

I've found an issue when exporting measurement as text. If you set the measurement as psychoacoustic in all SPL, and then export that as text, and select no smoothing. When you load that text file as a target curve it retains the no smoothed version. Is this a bug or just the new functionality?

Because what I used to do was smooth a measurement in all SPL first, then don't use smoothing in export as text because it used to double smooth. You'd get smoothing from the all SPL and the export as text. Now, when you export as text it no longer obeys all SPL smoothing, and appears to only obey the smoothing type that's chosen in export as text.

I'd just like to know if this is how it is gong forward or is this going to change? If this is new then I just need to set my export as text to psychoacoustic

 

[John Mulcahy]

I've found an issue when exporting measurement as text. If you set the measurement as psychoacoustic in all SPL, and then export that as text, and select no smoothing. When you load that text file as a target curve it retains the no smoothed version. Is this a bug or just the new functionality?

Because what I used to do was smooth a measurement in all SPL first, then don't use smoothing in export as text because it used to double smooth. You'd get smoothing from the all SPL and the export as text. Now, when you export as text it no longer obeys all SPL smoothing, and appears to only obey the smoothing type that's chosen in export as text.

That's not new behaviour and there was never double smoothing. The options on the export dialog are to either export the measurement as it is, with whatever smoothing it has, or to export a version with your chosen smoothing.

 

[John Mulcahy]

Builds updated today (beta 21) with these changes:

• Added: Export measurement as text has an option to omit phase
• Added: API endpoints to access the IR window settings for each measurement
• Fixed: Stepped level THD measurements could be misinterpreted as stepped frequency
• Fixed: Exporting a log spaced measurement as text could have zero as the final SPL value
• Fixed: The notes for a measurement made with filters had an extra "Hz" after each frequency
• Fixed: When using RME interfaces on macOS starting audio capture would fail if the generator was already running
• Fixed: Capturing a graph image would fail if the comment was selected to be on the image
• Fixed: Division with regularisation could produce an empty result

 

[HD MOVIE SOURCE]

That's not new behaviour and there was never double smoothing. The options on the export dialog are to either export the measurement as it is, with whatever smoothing it has, or to export a version with your chosen smoothing.

Okay, maybe I got my wired crossed. I never apply smoothing after a calculation has been done in REW, because the result is already smoothed. I thought it was like that all the time. I didn't realize that the functionality was different for actual measured responses, vs calculated responses that have been smoothed. I understand now.

Example, harmon target over smoothed measurement = a over b mag response that is smoothed. I export that with no smoothing and it's perfect. I just didn't realize measurements weren't like that too.
Thanks.