OpenViBE forum

Hi,

I've been trying to use the Emotiv EPOC headset along with OpenVibe to Spell words by the P300 Speller co-adapt scenario.

I do the calibration phase and the signal obtained is quite good. I run the train xDawn and the train classifier and both seems to work properly (they stop automatically, classifier has almost 90% accuracy).

However, when I try to use the copy mode or the free spelling mode, the results are pretty bad. Basically the scenario is not able to identify any letter. Sometimes the system identifies the wrong letter, and sometimes the system does not identify any letter at all (none response are given).

I've been trying to improve the calibration phase, making it longer, but it has not improving the accuracy. Furthermore, the signal obtained (analysing the image) is good enough. The problems seems to be something related to the classifier, as if it is not working. I also tried to vary the CoAdaptP300_stop_condition, but it does not worked as well.

I don't know what else can I do!
I would appreciate your help!

Thank you very much.

Artur Quirino

Hello

The P300 Speller coadapt relies on shared memory tagging to send event markers. We have noticed a time drift of the tagging, which degrades the performances of the classifier over time.

We are about to release a new version of the P300 Speller which corrects this issue.

In the meantime I recommend you to use the previous version of the P300 Speller which has been reported to work with the Emotiv EPOC.

You'll find a tutorial on this webpage: http://openvibe.inria.fr/p300-speller-xdawn/

Best regards

nfoy wrote:The P300 Speller coadapt relies on shared memory tagging to send event markers. We have noticed a time drift of the tagging, which degrades the performances of the classifier over time.

We are about to release a new version of the P300 Speller which corrects this issue.

Nathanael, hi.

It sounds like the new version of the CoAdapt speller will be using the TCP tagging for event markers, is that correct? Are there any beta test versions available? We've had some questions on this at the OpenBCI forum. http://openbci.com/forum/index.php?p=/d ... ons#latest

On the OpenViBE front page a Christmas release for v1.3 is mentioned. Is that a pre-requisite for the new CoAdapt?

Regards,

William Croft, OpenBCI

Hello William,

we don't currently have the software engineering resources to maintain the CoAdapt P300 speller functional as part of OpenViBE at Inria, which is the reason why its disabled at the moment. Team Athena at Inria Sophia-Antipolis has developed a completely rewritten version of it, but merging it with OpenViBE is not presently on the todo-list afaik. You could try contacting Athena about it.

Edit: The rewritten version of CoAdapt P300 is pending on some paper publications getting through.

At the moment the 'basic' and xdawn P300 spellers should be functional in OpenViBE v1.2.x using the TCP (software) Tagging, although they're likely slower than the CoAdapt one.

If anybody patches the 0.18 CoAdapt P300 to work with 1.2.x, we're happy to consider merging the patch. We're not working on any P300 related advances for 1.3 presently; most of the OV engineering resources are currently dedicated to the medical certification project of the software core.


Best,
Jussi