After much too long, our article (with @UniBremenCSL @emilymugler @ASPENLABVCU @MattGoldrick and @SlutzkyLab ) is finally out:
"Generating Natural, Intelligible Speech From Brain Activity in Motor, Premotor, and Inferior Frontal Cortices"https://www.frontiersin.org/articles/10.3389/fnins.2019.01267/full …
-
-
Our approach is based on Unit Selection, where the best fitting unit of speech is just concatenated to the output. We determine the best fitting unit of speech by comparing the cosine similarity of high-gamma activity. This video illustrates the approach: https://www.youtube.com/watch?v=GDsb99rDDgo …pic.twitter.com/7D1eSoTec6
Prikaži ovu nit -
This simple and fast approach reconstructs very high quality audio using the patient's own voice. You can listen to a few examples in the supplementary material.pic.twitter.com/HorVv3RtI4
Prikaži ovu nit -
The approach used also allows us to run the analysis with the three regions individually. Each of them performed above chance level, but M1 clearly provided most information for the decoding processes.pic.twitter.com/CJTgvf8QpE
Prikaži ovu nit -
We also asked volunteers to perform a listening test and 66.7% of the reconstructed words could be identified correctly in a forced-choice test with 4 options.pic.twitter.com/SQm18pDVni
Prikaži ovu nit -
What I like best about this paper is that this super simple pattern matching approach is able to produce high-quality audio, while being fast enough for real-time processing at the same time. It was great fun working with everyone involved in this project. Thanks!
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.