This repository contains the experiments carried out for the project topic Self-label correction for noisy labels in Speech Emotion Recognition.
pip install requirements.txt
added a PDF
Dataset creation is described in detail in section 3.1. IEMOCAP dataset was obtained by following the instructions on this website.
data-explore.ipynb contains information about the creation of noisy datasets.
To run experiment I, edit config.py according to the settings mentioned in section. Edit the train paths in train-pytorch.sh to train the appropriate models. For training, hard-26.48-train.csv is used for noisy-A and hard-40.19-train.csv is used for noisy-B models. Use the --noisy option when training with noisy labels instead of clean labels.
For testing, use test.sh. You can use either hard-26.48-test.csv or hard-40.19-test.csv (both are the same) as test paths.
To validate the ML-algorithms experiment, use train-sklearn.sh with feature_name = handcrafted.
results_analysis.ipynb contains the code to reproduce the classification report and confusion matrices for the given config
Set an appropritate run_name in config.py for all experiment runs.
All experiments use wandb for tracking, turning it to offline mode will default to not saving any model.
To run experiments with ProSelfLC, change loss_type under config.py to proselflc.
Use results_analysis.ipynb to reproduce the results as mentioned previously.
Use pretty_plots.ipynb to generate the line graphs mentioned in section 4.2.1 and 4.2.3. These require the CSV logs from wandb cloud.
Use results_analysis.ipynb to generate the gender-specific confusion matrices and classification results.
Each train run saves a file called train-predictions-<model-name>-<hyperparameter-setting>.csv file that contains information about labels that were flipped during training. Use results_analysis.ipynb to get the label pairs that were flipped sorted in descending order during the specific run.
Qiuqiang Kong, Yin Cao, Turab Iqbal, Yuxuan Wang, Wenwu Wang, and Mark D. Plumbley. "Panns: Large-scale pretrained audio neural networks for audio pattern recognition." IEEE/ACM Transactions on Audio, Speech, and Language Processing 28 (2020): 2880-2894.
Wang, X., Hua, Y., Kodirov, E., Clifton, D.A., and Robertson, N.M. "Proselflc: Progressive self label correction for training robust deep neural networks." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 752-761, 2021.