Speech Prosthesis: Problems and Solutions
Background: Locked-in patients with intact speech motor cortices continue to have neural signals that encode their speech. Recording long-term from these signals and decoding their patterns of firings should provide a basis for speech prosthesis. Studies with the Neurotrophic Electrode (NE) implanted in five subjects strongly suggest that long-term recordings of stable signals are now feasible. In one subject implanted in the speech motor cortex, decoded signals have allowed him to produce phonemes.
Design/Methods:The Neurotrophic Electrode is designed as a glass, hollow-tipped electrode containing 2 mil teflon insulated gold wires. Trophic factors attract neurites into the tip where they become myelinated and grow through the tip as a bridge that securely holds the electrode tip in place, hence resulting in stable long term recordings. The wires provide bipolar recordings. Implantation within the speech motor cortex allows recordings from neurons controlling the articulators of jaw, tongue and face. Thus the task becomes a motor task of detecting the signals associated with the movements of the articulators.
Results:The present subject, ER, was implanted in 2004 with a single three-wire Neurotrophic Electrode and fully implantable electronic package. Over the years, phonemes have been identified by the firing patterns of 40 stable recorded neurons using linear discriminant analysis, peri-event time histograms, neural net and other decoding paradigms. Additionally, in a 2D formant frequency task, he was able to move the cursor over a 2D space that contained the vowels, 'uh', 'ih', 'iy', 'oo', and 'ah'. Beginning with the center 'uh' vowel phoneme, he could move the cursor to the position of the other vowels with 80% accuracy.
Conclusions:These results strongly suggest that speech prostheses are now feasible. We intend to take the identified phonemes, concatenate them into short phrases of at least 500 ms (because that is the minimum segment for decoding) and present these to the subject. The decoded neural signals should provide output of these phrases through the computer’s speakers. To record multiple signals in humans, an improved multi-channel Neurotrophic electrode and multi-channel implantable electronics package are required.
Study Supported by: NIDCD, NINDS.