PP20-21_encoding_model

The code was created as part of a Practical Project at the Applied Neurocognitive Psychology Lab/University of Oldenburg in 2020 - 2021 under the supervision of Moritz Boos and Dr. Arkan Al-Zubaidi. The code provides an encoding model to predict fMRI data stored in BIDS format from natural speech.

1. Extracting the Modulation Power Spectrum: This function extracts the Modulation Power Spectrum from an auditory stimulus in a BIDS compliant format. The function is based on the MEL spectrogram and the 2D Fourier Transform. Use with python wav_files_to_bids_tsv_2.py path/to/your/wavfiles/*.wav -c path/to/your/config.json Extracting MPS. By default extracts the MPS. For a fully commented script and step by step explanation please refer to the step-by-step MPS Extraction here.

2. Validate the model: Use the Voxelwiseencoding App by Moritz Boos (for the script, please refer to Validating Voxelwiseencoding Model, for a step by step application, see step-by-step Validating VWE)

Usage for extracting the MPS

 
positional arguments:
-------
  filename :      str, path to wav files to be converted. Can be used with wildcard *.wav. 

keyword arguments:
------
  sr:             int, sampling rate of auditory files (samples per second: 44100 Hz by default)
  n_fft:          int, window length of spectrogram (default 882)
  mps_n_fft:      int, window length for extracting the MPS (default 100)
  hop_length:     int, step size for extracting MEL spectrogram (default 882)
  mps_hop_length: int, step size for extracting MPS (default 100)
  n_mels:         int, number of mels used (default 64)
  
  
optional arguments:
------
  plot_mps:       bool, plotting the mel spectrogram and mps forthe first window side by side (by default set to True)

Output

The function returns three outputs:

1. a representation of a feature matrix of shape (samples x features)
2. Stimulus Repetition Time (int)
3. the names of the features (as list of strings)

Optional

The function can return the plotted MPS. By default, this is set to True and has to be indicated if needed otherwise.

Note: Default settings are set so the windows for the extraction of the Mel spectrogram and the MPS each are non-overlapping.