Batch Reject and other issues about deepsqueak

[DrCoffey]

Hey Marek,

Tonality thresholds can be adjusted using the slider next to the tonality graph. It will effect the tonality threshold for all calls in the files.

I think the batch reject is off the screen because you have a low resolution monitor. Russell is working on a little fix to make sure the window will show up on any screen.

You can modify the CalculateStats.m code if you want to calculate different statistics, but you would need to know how to code in matlab. You would also likely need to modify the excel_Callback.m.

We are definitely going to release a standalone version of DeepSqueak very soon, but it won't receive the support or updates that the main branch does.

-Kevin

[MxMarx]

Hi Marek,

The batch reject by threshold window should now appear on the screen.
Were you getting a specific error from batch reject? The error would appear in the Matlab command window.

I've compiled DeepSqueak into an executable.
I've tested it on my computer running Windows 10 and detection and clustering all seemed to work.

What particular parameters would you like to extract? The way to do it would be to calculate them in CalculateStats.m, which excel_Callback uses to create the output.

-Russell

[Marrekko]

Hey Kevin, hey Russel,

thank you for your fast answer! Before updating the new files i checked for error messages (was not the case) and tried different resolutions than my 1920x1080 (didnt work). However, with the update the prompt shows up correct and batch reject works fine now, so thanks a lot for this very quick problem solving!

Also, thank you for the executable standalone! I will for sure test it next week, really good job again! But as you said it wont get as much support as your main branch, is it general possible to build a standalone with new updates by ourself using some python tools?

We are not sure which parameters we will need in future but knowing which files have to be edited is definitely big step forward, thank you!

Tonality thresholds can be adjusted using the slider next to the tonality graph. It will effect the tonality threshold for all calls in the files.

Actually, i was totally unaware of that. Of course i know the slider but i didnt knew it changes tonality for all calls. What is the reason for that and is there a way to bring this feature up for individual calls? In general the automatic tonality setting works pretty good, but i found it necessary to adjust it for some cases to get a proper fitting contour and therefor extracted statistics of the call.

Thank you so much again for your superb support!

Marek

[VoiceScientist]

I second the request for adjusting tonality for individual calls. We have the same issue with needing to adjust the setting for some calls to improve the contour, but not all - mostly it depends on the SNR of individual calls. I also echo Marek's appreciation of your superb support!
Aaron

[DrCoffey]

Hey Aaron & Marek,

I see the value of adjusting tonality of individual calls, but it seems difficult to implement. We tend to use DeepSqueak on huge data sets (10's of thousands of calls). Adjusting tonality for all calls is basically impossible, so we have to rely on a global tonailty setting. I'm also not sure how we could code this, we would need to retain global tonailty thresholds for most calls, and then remember for each call if it was changed manually. I will think about it, but I'm not sure we can do it.

Unfortunately we are running out of time to work on DeepSqueak, so we will not be doing many other feature changes. We will continue to try and implement bug fixes however. We are hoping that the community will start coding in features that they really want and send us the pull requests to test and integrate.

-Kevin

[Marrekko]

Dear Kevin and Russel,
I hope you are good and you still have a little bit of time for DS. Sorry for coming back to this issue, but whilst analyzing a new data set I found a bug in the Windows standalone version. This regards training of the denoising Network, such to say a core function of DS, and was replicated on several machines.

Error using trainNetwork (line 150)
Deployment of training progress plot not supported.

Error in TrainPostHocDenoiser_Callback (line 143)

Error in gui_mainfcn (line 95)

Error in DeepSqueak (line 30)

Error in matlab.graphics.internal.figfile.FigFile/read>@(hObject,eventdata)DeepSqueak('TrainPostHocDenoiser_Callback',hObject,eventdata,guidata(hObject))

Error using waitforallfiguresclosed (line 9)
Error while evaluating Menu Callback.

Obviously, the Matlab Compiler Runtime has problems to generate figures (a common problem with compiled scripts). As I dont find those figures to be necessary I would change code to not generate those. However, the Matlab compiler is a product we don't have access to. Do you have any solution for this, or find a way to change the compiled Windows version with less effort?

Best
Marek

[MxMarx]

Hi Marek,

I've recompiled everything with the newest version of Matlab, hopefully that fixed the issue.
If it doesn't let me know and I'll make it not show the plot in the compiled version.

[Marrekko]

Hi Russel,

thank you so much for the work and the effort! I'm very sorry but I just tried it and it seems to have the same problem. Everything else seems to work fine as in the previous version.

Best
Marek

[bergagi1]

Hello MxMarx,

thanks for the previous standalone versions.
Are you planning in doing smt similar for DS V3?

Best,
Giorgio