PMID-20011034[0] A Wireless Brain-Machine Interface for Real-Time Speech Synthesis
- Neurites grow into the glass electrode over the course of 3-4 months; the signals and neurons are henceforth stable, at least for the period prior publication (>4 years).
- Used an FM modulator to send out the broadband neural signal; powered the implanted electronics inductively.
- Sorted 56 spike clusters (!!)
- quote: "We chose to err on the side of overestimating the number of clusters in our BMI since our Kalman filter decoding technique is somewhat robust to noisy inputs, whereas a stricter criterion for cluster definition might leave out information-carrying spike clusters."
- 27 units on one wire and 29 on the other.
- Quote: "neurons in the implanted region of left ventral premotor cortex represent intended speech sounds in terms of formant frequency trajectories, and projections from these neurons to primary motor cortex transform the intended formant trajectories into motor commands to the speech articulators."
- Thus speech can be represented as a trajectory through formant space.
- plus there are many simple low-load formant-based sw synthesizers
- Used supervised methods (ridge regression), where the user was asked to imagine making vowel sounds mimicking what he heard.
- only used the first 2 vowel formants; hence 2D task.
- Supervised from 8 ~1-minute recording sessions.
- 25 real-time feedback sessions over 5 months -- not much training time, why?
- Video looks alright.
____References____
[0] Guenther FH, Brumberg JS, Wright EJ, Nieto-Castanon A, Tourville JA, Panko M, Law R, Siebert SA, Bartels JL, Andreasen DS, Ehirim P, Mao H, Kennedy PR, A wireless brain-machine interface for real-time speech synthesis.PLoS One 4:12, e8218 (2009 Dec 9) |
|