Experiment based on the P300-speller

[stefan]

Hello everyone ,

I'd like to thank OpenViBE team for their amazing software
I’m doing an experiment based on the P300-speller with Emotive headset, and I’ve few questions:

1. Do I need to make changes to the available scenarios on OpenViBE to be employed using Emotive set?

2. Is it possible to replace the letters on ‘P300-speller visualization’ with pictures? And how can I do this? If not how can I build similar board visualization with pictures instead of letters? I have P300-speller.ui instead of p300-speller.glade?
Is it possible to attach sounds to each picture? And how can I reach this?
I know about the magic card but it has different characteristics which are not applicable to my experiment.

3. How the available p300-speller scenario can be used? What is the logical order? How many repetitions are required to train the classifier? and how can I control the player if more than one is involved, specifically regarding training the classifier?

4.What is the reason behind the unordered visual evoked? Why the scanning is not running on order, e.g. scanning the rows from the top to the bottom, or scanning the columns from the left to the right?


Thank you.

[jlegeny]

Hello stefan,

thank you for using OpenViBE. I'll try to answer your questions to my best

1. If you use the xDAWN scenarios then, in theory, there is nothing to modify inside. The xDAWN spatial filter will select the best electrodes on its own.

2. It is possible, you will have to edit the .ui file in Glade Interface Designer and replace the labels (containing the letters) in eventboxes by images. As for the sound, what exactly do you mean by attaching it to each picture ?

3. The scenarios in the p300-speller group should be used in order (as numbered), for more information about these look into our documentation: http://openvibe.inria.fr/documentation/ ... XDAWN.html The number of repetitions depends on the precision you want to achieve and of the headset you are using. You will probably need more repetitions with Emotiv than with a medical EEG. What do you mean by involving more players ?

4. The problem is that if you flash two targets which are too close successively, they might both trigger the p300. In ideal case, you want to flash targets which are as far away as possible every time.

[stefan]

Hello Jlegeny,

Thank you for your prompt replay and support

About the sounds: I intended to add a voice for each letter or picture. For example, if the user wants to spell the letter A, then a voice says ‘A’ will be played when this letter is chosen, for providing audio feedback. Is it possible to do so with OpenViBE?

For the players: I am using 8 users on my experiment, but I want to process each personalized data individually and do not want to mix the users’ data. Can I simply close the software and reopen it for the new user to delete the training data available from the past user?

Sorry for asking many questions but I’ve some more

A. Is there a detailed documentation for the available scenarios? To describe the path of the scenarios. For, example, I do not understand why the training scenario is very different of the online scenario for the P300-speller? And what is the meaning behind these differences? Can you provide a detailed explanation of the use of ‘identity’ box on this scenario and the following ramification?

B. I would like to record the EEG data produced during the training and online phase of the speller. Is the only thing I need is replacing the ‘Generic stream reader’ box with the ‘acquisition client’ box where ever it accrues? How about the connections?

C. Are the massages the only way to visualize the results? Or there are other ways to extract the results?

Many thanks

[jlegeny]

Hi Stefan,

About the sounds: I intended to add a voice for each letter or picture. For example, if the user wants to spell the letter A, then a voice says ‘A’ will be played when this letter is chosen, for providing audio feedback. Is it possible to do so with OpenViBE?

OpenViBE provides a box which plays a selected sound when it receives a stimulation. The problem is that you need one box for each sound, for speller this would mean adding a lot of boxes to the scenario. Also you would need some logical processing to convert target row and target to a single stimulation. You can, however use the sounds quite easily with the magic card scenario.

For the players: I am using 8 users on my experiment, but I want to process each personalized data individually and do not want to mix the users’ data. Can I simply close the software and reopen it for the new user to delete the training data available from the past user?

To answer your second question, answering your other questions might help.

Sorry for asking many questions but I’ve some more :roll:

> A. Is there a detailed documentation for the available scenarios? To describe the path of the scenarios. For, example, I do not understand why the training scenario is very different of the online scenario for the P300-speller? And what is the meaning behind these differences? Can you provide a detailed explanation of the use of ‘identity’ box on this scenario and the following ramification?

The documentation of the existing boxes can be found here : http://openvibe.inria.fr/documentation/ ... nario.html

Identity box does nothing actually, it is used only to make the scenario cleaner. In order to better comprehend the working of a p300 speller I would recommend reading some research papers on the subject. In short, the training scenarios only record and flag signals during the first acquisition. After that you run training scenarios which will generate a classifier which can find a p300 peak. In the online scenario this classifier is used to detect peaks in every line and columnt at the moment of the flash.

B. I would like to record the EEG data produced during the training and online phase of the speller. Is the only thing I need is replacing the ‘Generic stream reader’ box with the ‘acquisition client’ box where ever it accrues? How about the connections?

To get hold of the basic functionnalities of OpenViBE I would recommend watching our tutorial video http://openvibe.inria.fr/video.php?q=vi ... ylist=true

EEG data itself is produced only in the acquisition and online test scenarios. This data is already recorded by the generic stream writer box. In order for the p300 scenario to work you have to run the acquisition scenario first, then open the training scenario and change the file used by the generic stream reader to the one recorded in the previous step. You have to do this for each subject, this is why there is no reason to restart openvibe, as the only product of the acquisition scenario is the EEG file.

C. Are the massages the only way to visualize the results? Or there are other ways to extract the results?

This depends on results you wish to visualize. There are boxes which enable you to see the performances of a classifier visually.

Cheers
Jozef