Brain state optimization with audio stimuli
A method and system for generating an optimal audio stimulus for achieving a target brain state value for a brain state. The method and system can be used to generate one or more brain state models which can decode brain activity signals to predict brain state values. The brain state models can be applied to brain activity signals captured while users are performing tasks with an audio stimulus. Audio features of the audio stimulus can be extracted. An audio-brain model can be trained on the predicted brain state values and the audio features. From the trained audio-brain model, the optimal audio stimulus can be generated.
1 . A method for brain state optimization, the method comprising:
prompting a set of users to perform a sequence of tasks in coordination with provision of a stimulus;
recording a brain activity signal for each user of the set of users while the user is performing the sequence of tasks;
receiving a reported brain state value, reported by each user of the set of users, for each task of the sequence of tasks;
training a first brain state model based on the brain activity signals recorded from the set of users and the reported brain state values, wherein the first brain state model is configured to process an input brain activity signal and to output a predicted brain state value based on the input brain activity signal;
applying the first brain state model to predict a first brain state value for each user and each task;
extracting a set of features from the stimulus;
training a stimulus-brain model based on the set of features and the first brain state values over the set of users and the sequence of tasks; and
generating an optimal stimulus to achieve a target brain state value based on the trained stimulus-brain model.
2 . The method of claim 1 , wherein the sequence of tasks includes a first task from a first category of tasks and a second task from a second category of tasks different from the first category.
3 . The method of claim 1 , wherein the brain activity signal recorded for each user of the set of users comprises a plurality of electroencephalogram (EEG) channels.
4 . The method of claim 1 , further comprising:
partitioning the brain activity signals for the set of users into brain activity epochs using a sliding window; and
featurizing the brain activity epochs, wherein the brain state model is trained on the features for the brain activity epochs.
5 . The method of claim 1 , wherein the first brain state model is one of: a regression model, a neural network, a random forest model, and a decision tree model.
6 . The method of claim 1 , wherein the first brain state model is trained on the brain activity signals from a random subset of users from the set of users.
7 . The method of claim 6 , further comprising:
training a second brain state model based on the brain activity signals from a second random subset of users from the set of users, wherein the second brain state model is configured to input a brain activity signal and to output a predicted brain state value based on the input brain activity signal;
for each user and each task, applying the second brain state model to the brain activity signal to predict a second brain state value; and
wherein training of the stimulus-brain model is further based on the second brain state values over the set of users and the sequence of tasks.
8 . The method of claim 7 , further comprising:
for each user, selecting between the first brain state values and the second brain state values that are closest to the reported brain state values over the sequence of tasks for the user; and
wherein the training of the stimulus-brain model is based on the selection.
9 . The method of claim 1 , wherein extracting the set of features from the stimulus comprises:
partitioning an audio stimulus into audio epochs using a second sliding window; and
extracting the set of features for each of the audio epochs.
10 . The method of claim 1 , further comprising:
prompting the set of users to reperform the sequence of tasks in coordination with provision of a second stimulus;
wherein recording the brain activity signals comprises recording while the set of users are reperforming the sequence of tasks in coordination with the second stimulus, and
wherein the set of features is also extracted from the second stimulus, for training of the stimulus-brain model.
11 . The method of claim 1 , wherein generating the optimal stimulus further comprises optimizing for a second brain state.