Sound Localisation Error

[Dip]

I am trying to do sound localisation using ps-eye. Since it is the supported hardware by the hark, i downloaded the transfer function for sound localisation and separation from the hark website. I created a network in the hark designer as explained in the hark cookbook but i have been getting errors regarding the a_matrix parameter of localise_music since a couple of days. I ran through faq and found that we have to zip the files of transfer fucntion downloaded from the website (tsp recording) to be supported in the network. Hence i did that. Still, i am getting these kind of errors.

“Stopped
UINodeRepository::Scan()
Scanning def /usr/lib/flowdesigner/toolbox
done loading def files
loading XML document from memory
done!
Building network :MAIN

[E] [zip_open failed: file ‘localization’ unexpected error] [harkio_TransferFunction_fromFile] [166]
LocalizationMUSIC. Failed to read tf file.
reading A matrix

Any idea where I went wrong?

Attachments:

1. 

   Screenshot-from-2017-10-27-16-38-22.png
   

2. 

   Screenshot-from-2017-10-27-16-37-46.png
   

[lag]

You need to specify the transfer function zip file you got from the website in the A_MATRIX setting : ex : /home/your_user/Downloads/pseye_rectf.zip

I’m also trying out hark with a pseye and would be interested in your results.

Good luck.

[Dip]

Yes. Following the demo pack hark cookbook suggested, I figured that out.

That is really great to know. I could get some help out here. I am a beginner in this field.

I used the demo network from here http://www.hark.jp/networks/HARK_recog_2.3.0.1_practice2.zip , changed it to suit my pseye. It shows sound localization and separation results. But i am still figuring out to interpret those results because in sound localization graph, I get too many lines of graph even though speakers are just two. Plus, the amplitude of separated sound is extremely low. I cannot hear them out properly. Has such problem occur to you? Is there gain or such parameters in it? I couldn’t find.

Anyway, I will be working on it whole week. Will be glad to know your results and get help, if i get stuck somewhere.

Regards.

[lag]

I only starting experimenting with hark yesterday but as of now, I haven’t been able to get decent result with the provided transfer functions.

I have tested capture with a ps eye and a kinect v1 but I havent been able to do precise localization or separation : it picks up noise a sound and it fails to pick up high powered voice…

I’ve had good results with offline samples from sample-hark-2_1_0 archive. The sample file was recorded using a kinect but when I try online with my kinect it does not work.

I suspect my room acoustic is too different from the room in which the transfer function was generated so I’m currently setting up a rig to measure my own transfer function.

I’ll update with progress.

Best Regards.

[Dip]

Yes. The offline network provided in the demo code works perfectly, separating the sounds and localising them. The online is for tamago device, in my demofile. Anyway, changing it to pseye transfer fucntion, didn’t help much with online network version.

I tried out a lot of parameters change and redesigned the network myself with guidelines from cookbook but noise still persists in pseye. I dont know what to do next.

With standard device supported by HARK, I dont think we need to regenerate the transfer function.

I am planning to move onto ROS. Maybe I will get some success in there. I will try out again today with this though.

I’ll update you with my results.

Regards.

[Yuka Goto]

Hello
Thank you for using HARK.
For PlayStation Eye, I assume you have read http://hark.jp/wiki.cgi?page=SupportedHardware#p3
and downloaded Transfer Function at http://hark.jp/wiki.cgi?page=SupportedHardware#p10
If the transfer function by TSP recordings does not perform well with tuning the THRESHOLD, then please try the other one by Geometrically Calculation and tune the THRESHOLD again.
To tune the value of THRESHOLD, please do the following:
1. Turn DEBUG, the parameter of SourceTracker node, ON
2. Run HARK
3. Obtain the results
4. Set THRESHOLD, the parameter of SourceTracker node, to be the middle value from the results above
5. Turn DEBUG, the parameter of SourceTracker node, OFF

http://www.hark.jp/document/hark-document-en/subsec-SourceTracker.html
http://www.hark.jp/document/hark-document-en/subsec-HarkParamsDynReconf.html

If it’s still not working, then you need to generate your own transfer function.

Best,
HARK Support Team

• This reply was modified 6 years, 5 months ago by Yuka Goto.
• This reply was modified 6 years, 5 months ago by Yuka Goto.
• This reply was modified 6 years, 5 months ago by Yuka Goto.

[lag]

Hello Yuka,

Thank you for your answer. I did also try the provided geometric TF without success but the room I am testing has unconventionnal acoustics as I’m under a roof and the ceiling has this shape :

____
/    \
|    |

So I’m trying to generate my TF. The first try did not work but I suspect I didnt use a good enough speaker. I will try today with a better quality speaker for the tsp source.

Any tips regarding the volume/power I should play the tsp at ?

Thanks !

[Yuka Goto]

Hello lag,

Thank you for participating in HARK Forum.
I am not sure how unconventional it can be to use HARK in your place (sorry my poor imagination), but I do understand you have no choice but creating your own TF.

> Any tips regarding the volume/power I should play the tsp at ?

TSP needs to be louder than the sound source that you actually want to localize later using your own TF.
It shouldn’t be too loud, though.
You can check the TSP data with sound editor such as Audacity by viewing it in spectrum mode to see if the data are too loud or not.
If the data are too loud, then the reflection will appear too clear.

Best,

HARK Support Team

• This reply was modified 6 years, 5 months ago by Yuka Goto.

[Author]

Posts