/DL_Dialogue_act_classification

DL Lab Project - Given a subset of switchboard corpus, goal is to classify dialogue acts from Speech and Text data. We define a RNN-LSTM model for Text classification and CNN model for speech classification and then ensemble both model to output a stable and higher performance model

Primary LanguagePython

Directory structure

  • code

    • TextRNNLSTMModel.py (Text-RNN)
    • SpeechCNNModel (Speech-CNN)
    • SpeechAndTextCombined (RNN-CNN-ensemble)

  • results

    • Train results plot
    • Test output

Sample -random utterence list similarly ordered for Speech and text data

IMS Directory Structure -/mount/arbeitsdaten31/studenten1/deeplearning/2017/infy -train set (training batch) -test set (testing batch) -dev set (dev batch) -text (text inputs , glove) -model (saved model)

Instruction

-Speech file main function contains two lines calling parameter for two functions, one is commented out
-Ensemble file also follows the same

Dependency

-Keras
-Numpy
-Panda
-Gzip
-tensorflow
-matplotlib