Subvocalization Pins/Settings
brandongrifs
Berkeley, CA
We are attempting to use the Cyton board to record subvocalizations of words. The process we are using is based on this article: https://towardsdatascience.com/merging-with-ai-how-to-make-a-brain-computer-interface-to-communicate-with-google-using-keras-and-f9414c540a92. However, the board used in the article is a Ganglion and when attempting to recreate the setup, our data is railing at the max or min value at every step.
Basically, we attach the data pins to our face at various areas described in the article and the grounding pins to our ears (both using conductive paste and gold electrode cups). We then think the word we wish to subvocalize to get an EEG signal corresponding to a word, and have even tried actually saying the words aloud to get better readings. I am wondering if anyone else has attempted this with the Cyton board and has any tips on which pins to use? Or in general, if anyone knows how to debug data collection issues with the Cyton board (especially related to pin setup and/or grounding)? Any help would be greatly appreciated.
Comments
Brandon, hi.
You might already have seen the original MIT paper,
https://dam-prod.media.mit.edu/x/2018/03/23/p43-kapur_BRjFwE6.pdf
It's somewhat unclear to me what signals are being received. Are they sEMG, surface EMG? With Cyton if your signal has too much amplitude, such that it is railing, you can use the SDK command strings to set a different scale factor. The default scale is 24, which works well for EEG. But sometimes ECG, EMG work sets a lower scale factor,
https://docs.openbci.com/docs/02Cyton/CytonSDK#channel-setting-commands
I have not read the MIT paper in detail, but I'd suggest following their method of referencing and grounding.
Regards, William
Here is a link in the 'old' docs section, showing how the scale is adjusted using the GUI,
https://github.com/OpenBCI/Docs/blob/master/Tutorials/01-Cyton_Getting Started_Guide.md#4-launch-the-gui-and-adjust-your-channel-settings
With Python, you can do this with the SDK command strings, as mentioned in the previous post.
Wondering if @Billh might comment on this 'subvocalization' research. Bill, this must be sEMG right? It's unclear to me if electrodes can pickup "nerve impulses" at the skin surface as described in the initial thread link,
https://towardsdatascience.com/merging-with-ai-how-to-make-a-brain-computer-interface-to-communicate-with-google-using-keras-and-f9414c540a92
My impression is that there are some misconceptions in that article. However the MIT research this was based on, does mention sEMG multiple times.
https://dam-prod.media.mit.edu/x/2018/03/23/p43-kapur_BRjFwE6.pdf
Regards, William
The article says they are picking up nerve signals but the electrode placement suggests this is surface EMG. Surface EMG has been used to pick up subvocalization since the 1960's. Movements of larynx, tongue and jaw muscles are typically the ones that surface EMG detects. I'd add that buccal and mental muscle located electrodes would be picking up lip motion as well, but usually subvocalization is used to refer to muscle contractions that a lip reader could not see.
Bill, if this is indeed surface EMG, what is your impression of the initial link article's suggestion to place ground and reference at the ear lobes? That is NOT what the original MIT research used. Nor the photo shown from NASA below. Normally with EMG one places channel and reference both near the muscle being measured.
https://en.wikipedia.org/wiki/Subvocal_recognition
Images of typical subvocalization placements,
https://www.google.com/search?rlz=1C5CHFA_enUS577US577&tbm=isch&sa=1&q=semg+placement+subvocalization&oq=semg+placement+subvocalization
Ground and reference can certainly be further away-- I think putting reference further away might make amplitudes slightly higher, and it would not matter. You could try several locations for these and decide what worked best.