Direct Speech Synthesis from Non-Invasive, Neuromagnetic Signals

Jinuk Kwon^1,2, David Harwath³, Debadatta Dash², Paul Ferrari⁴, and Jun Wang^1,2

¹ Department of Speech, Language, and Hearing Sciences, The University of Texas at Austin, USA
² Department of Neurology, Dell Medical School, The University of Texas at Austin, USA
³ Department of Computer Science, The University of Texas at Austin, USA
⁴ Jack H. Miller Magnetoencephalography Center, Helen DeVos Children’s Medical Center, USA

 Target: Audio synthesized using the target Mel spectrogram     

Generated: Audio synthesized using the Generated Mel spectrogram from MEG     

      Phrase 1: Do you under stand me?

      Phrase 2: That's perfect

      Phrase 3: How are you?

      Phrase 4: Good-Bye

      Phrase 5: I need help