How to reproduce P300 Speller using Emotiv EPOC headset

[davidchung]

Hi all,

Hurray! For the first time, I was able to achieve 100% correct result (e.g. 91I59NI2RS) even though my Classifier performance was approx. 78% after the training phase.

Question 1.
What happens if I run/play multiple times in the "p300-speller-2-train-xDAWN.xml" (or "p300-speller-3-train-classifier") in the training phase? Does the new training gets accumulated and stored in the .ov file OR does it overwrite the previous training in the .ov file?


Question 2.
I would like to change the dimension of the P300 matrix. Also I would like to change the graphics? For example, 3 x 3 matrix with photos. Is there an easy way to do this?


Question 3.
If I had to remove some of the EEG sensors of the Emotiv EPOC headset to minimize the total number of sensors, which EEG sensors should I keep for P300-speller in the Emotiv?

Channel names in Emotiv headset based on the International 10-20 locations are: AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4


Question 4.
Any plans for OpenViBE technical conference? Or any recommendation for upcoming good BCI conferences?

Thank you so much!
David

[yrenard]

Dear davidchung,

congratulations for getting this to work ! 78% is not an excellent classifier performance in case of the P300 speller but with 12 repetitions, it indeed leads to correct detection. Did you try to play with the number of repetitions as well ?

Question 1.
What happens if I run/play multiple times in the "p300-speller-2-train-xDAWN.xml" (or "p300-speller-3-train-classifier") in the training phase? Does the new training gets accumulated and stored in the .ov file OR does it overwrite the previous training in the .ov file?

The training is not iterative. The data is recorded in separate files so you can train on any of them but not all of them. If you want more data to train on, just spell more letters in one session.

Question 2.
I would like to change the dimension of the P300 matrix. Also I would like to change the graphics? For example, 3 x 3 matrix with photos. Is there an easy way to do this?

Did you look at the Magic Card scenarios ?

Question 3.
If I had to remove some of the EEG sensors of the Emotiv EPOC headset to minimize the total number of sensors, which EEG sensors should I keep for P300-speller in the Emotiv?

Channel names in Emotiv headset based on the International 10-20 locations are: AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4

I recommand you read this article http://hal.archives-ouvertes.fr/hal-00517388

Question 4.
Any plans for OpenViBE technical conference? Or any recommendation for upcoming good BCI conferences?

There will be some OpenViBE workshop by the end of 2011 but not before this. If interested in BCI events, you should have a look at Fabien Lotte's dedicated page.

Hope this helps,
Yann

[davidchung]

Question 2.
I would like to change the dimension of the P300 matrix. Also I would like to change the graphics? For example, 3 x 3 matrix with photos. Is there an easy way to do this?

Did you look at the Magic Card scenarios ?
Yann

I will create a new thread for this.

[davidchung]

Hi

I am not sure how "p300-speller-4-online.xml" knows which training set to choose, if there are multiple "p300-speller-xdawn-4-[yyyy.mm.dd-hh.mm.ss].ov" files under \openvibe\dist\share\openvibe-scenarios\signals.

Thanks,
David

[lbonnet]

Hi David,

I am not sure how "p300-speller-4-online.xml" knows which training set to choose, if there are multiple "p300-speller-xdawn-4-[yyyy.mm.dd-hh.mm.ss].ov" files under \openvibe\dist\share\openvibe-scenarios\signals.

The online scenario does not directly use the brain signal, but the configuration files generated during the training phase on this signal. These files are :
- openvibe-scenarios/bci/p300-speller-xDAWN/p300-spatial-filter.cfg for the xDAWN spatial filter box
- openvibe-scenarios/bci/p300-speller-xDAWN/p300-classifier-processor.cfg for each classifier processor box.

Every time your run the training scenarios (xDAWN and classifier) you generate new configuration files. The name of these files never change (it's the name provided in the classifier trainer box and the xDAWN spatial filter trainer box)

regards,
Laurent

[williamblake621]

which one help to indicate brain for working or any movement ?

[zackbam]

Hello!
I am using the p300-speller-xDAWN scenario, trying to get some P300 using Emotiv. The 10-fold validation accuracy estimation I get is between 66-72%. When I try the online scenario, in the best case I get 1/20 letters detected. Really poor performance (almost random). I tried increasing the repetitions by changing the values of p300-speller-stimulator.cfg file, but nothing was improved.
Apart from changing the source filename in the xDawn training and the classifier training scenarios, is there something else I should change to improve performance with Emotiv? davidchung and yrenard, am I doing sth wrong?

[tshanno]

Hello!
I am using the p300-speller-xDAWN scenario, trying to get some P300 using Emotiv. The 10-fold validation accuracy estimation I get is between 66-72%. When I try the online scenario, in the best case I get 1/20 letters detected. Really poor performance (almost random). I tried increasing the repetitions by changing the values of p300-speller-stimulator.cfg file, but nothing was improved.
Apart from changing the source filename in the xDawn training and the classifier training scenarios, is there something else I should change to improve performance with Emotiv? davidchung and yrenard, am I doing sth wrong?

Did anyone ever answer this? I am experiencing similar problems, however, my validation accuracy is 82%.

Thanks!
Tara

[jtlindgren]

Hi Tara,

one of our interns has successfully used xdawn p300 with Emotiv Epoc and openvibe 1.2.2 recently, so it should be doable. The accuracies should in general be better than those obtained in 2012 and earlier, as we now have more precise alignment of the markers to the eeg signal.

Unfortunately its difficult to debug the p300 apart from the trivial stuff like looking that the signal looks correct before starting the p300 session. Any appropriate debugging would involve testing with recorded and artificial data and analyze the results of different processing stages. This takes a lot of time - I've done such work in the past to test the p300 with gusbamp, but there are no easy recipes for such work.


Cheers,
Jussi