Figure 1.

Predictive modeling overview. The general framework of predictive models consists of three steps. First, input and output data are collected, for example during passive listening to natural speech sentences. Next, features are extracted. Traditional features for the neural activity can be the time-varying power of various frequency bins, such as high frequency range (70–150 Hz, shown above). For auditory stimuli, the audio envelope or spectrogram are often used. Finally, the data are split into a training and test set. The training set is used to fit the model, and the test set is used to calculate predictive score of the model.