Speech Emotion Recognition using Semantic Information

This repository provides training and evaluation code for the paper Speech Emotion Recognition using Semantic Information (ICASSP 2021). If you use this codebase in your experiments please cite:

Tzirakis, P., Nguyen, A., Zafeiriou, S., & Schuller, B. W. (2021). Speech Emotion Recognition using Semantic Information. arXiv preprint arXiv:2103.02993.

This repository provides the following:

1. Word2Vec embeddings trained on the German SWC corpus.
2. Speech2Vec embeddings trained on the SEWA DB.
3. Code to train/evaluate word2vec and our unified model, i. e., paralinguistic and semantic feature extrators with a LSTM cell on top.

Requirements

Below are listed the required modules to run the code.

• aeneas
• librosa
• nltk
• numpy
• stop-words
• tensorflow
• torch

Steps

1. Create the speech2vec segmentation by running speech2word_mapping.py in speech2vec folder.
2. Run data_generator.py to create tfrecords.
3. Run train.py to train the models, and eval.py to evaluate.