Fig. 1

Overview of the study paradigm, setup, and data analysis. a) The experiment consisted of three vocalization conditions: loud high pitch, loud low pitch, and soft high pitch. Subjects maintained a pitch in approximately 3 s intervals. Between these singing intervals, subjects could rest. The experiment was conducted in 60 s blocks (12 blocks in total), and, in each block, subjects performed one of the three singing conditions. Note that the definitions of high, low, loud, and soft were subjectively determined by each participant, and the musical notes shown in this figure are for illustration only and do not correspond to specific vocal frequencies. b) Larynx activity was recorded using 15 OPMs, 4 of which showed malfunctions. This resulted in 11 working OPMs (filled cubes in the schematic plot). c) Summary of data analysis. In each block, recorded signals were first filtered (high pass at 10 Hz and notched at 50 Hz and its harmonics up to 250 Hz). Then, the heartbeat artifact was removed using PCA. Next, signal envelopes were extracted using the Hilbert transform. We low-pass filtered these envelopes at 5 Hz. Next, we extracted intervals in which subjects sang based on the trigger signal (black rectangular trace) and concatenated all segments of the four blocks of each condition to construct condition matrices. Finally, we trained and tested an LDA classifier on these three matrices.