ser-uncertainty-quantification

Evaluating Uncertainty Quantification Approaches on Real-world Speech Emotion Recognition Tasks.

This repository contains all the code to reproduce the findings from the paper: Are you sure? Analysing Uncertainty Quantification Approaches for Real-world Speech Emotion Recognition

Getting Started

$ pip install -r requirements.txt

This project uses hydra to store the configuration of different configurations.

You can find the configurations under src\conf\experiments\[experiment name]

e.g. to run the experiments for the baseline based on Cross Entropy run:

$ cd src
$ python3 main.py +experiments=emo_cat_ce

Note: We do not have the rights to redistribute the MSP Podcast data set ourselves, therefore you must obtain it separately or exclude it from the training/test data.

License

This repository may only be used for non-commercial purposes (CC BY-NC-SA 4.0).

Citation

@inproceedings{schrufer24_interspeech,
  title     = {Are you sure? Analysing Uncertainty Quantification Approaches for Real-world Speech Emotion Recognition},
  author    = {Oliver Schrüfer and Manuel Milling and Felix Burkhardt and Florian Eyben and Björn Schuller},
  year      = {2024},
  booktitle = {Interspeech 2024},
  pages     = {3210--3214},
  doi       = {10.21437/Interspeech.2024-977},
}