PROJECT DESCRIPTION:

• Goal: Identify the importance of each electro-nodes in reconstructing human speech

• Data Source: Data is from an experiment that clinical doctor electro-nodes in a subject’s cortical surface and subject listens to 2100 human reading sentences. At the same time, recording audio signal and neural signal.

METHOD:

1. Segmenting neural signal by phones in each word in every sentence
2. Extract features from the segmented neural signal(ecog)
3. Doing classification on phones by regression on features extracted from neueral signal
4. Identify the most effective electro-nodes based on classfication error rate.

BUILD WITH:

• Data Manipulation & IO: pandas; numpy; h5py; pickle
• Parallel Computing: multiprocessing; joblib; functools
• Visulization: matplotlib, time
• Modeling: scikit-learn

HOW TO USE:

• Environment(IDE): Spyder
• Steps:

1. open Spyder
2. click "Projects" in the menu
3. then "Open Project..."
4. go to this README.md directory
5. click choose

CODE STRUCTRE:

• main.py (user interface)

• load.py (load all the data)

• function/DataManipulation.py (raw data scaling)

  • DataManipulation.py (ecog raw data manipulation)
  • SentenceSegmentation.py (ecog data segmentation by phones)
  • FeatureGeneration.py (create features from segmented ecog data)
  • Visualization.py (dimension reduced data visualization)
  • Classification.py (compared the classification performance)
  • util.py (utility functions)