Fig. 1
‘Rec-Dec-Back’ in the Nostalgia Brain-Music Interface (N-BMI). The N-BMI followed a three-step process called ‘Rec-Dec-Back,’ which refers to (1) recording, (2) decoding, and (3) feedback. During the recording step, participants listened to both self-selected and other-selected nostalgic songs and rated their subjective feelings of nostalgia, state-level well-being, and subjective memory vividness, while EEG activity was simultaneously recorded using an in-ear EEG device. In the decoding step, two models were developed: Model 1, the ‘Nostalgia Prediction Model,’ which utilized nostalgia ratings and the acoustic features of self-selected and other-selected songs extracted through the VGGish model; and Model 2, the ‘Nostalgia Decoder,’ which analyzed Fourier-transformed EEG power-frequency data recorded during song listening. Model 2 decoded nostalgic feelings by estimating whether the EEG patterns more closely matched those recorded during self-selected or other-selected song listening. In the feedback step, Model 1 predicted the participant’s nostalgic feelings and recommended a song from a pool of 7,481 songs to either enhance or reduce nostalgia. While the participant listened to the recommended song, Model 2 decoded nostalgic feelings, and both the decoded nostalgia and the song’s acoustic features were used to update Model 1. The updated Model 1 then predicted nostalgic feelings again and recommended another song from the pool. After listening to the recommended songs, participants rated their nostalgic feelings, state-level well-being, and subjective memory vividness.
