AFTER

Official implementation for the paper [Active Learning Based Fine-Tuning for Speech Emotion Recognition]

Dependencies

Python 3.9.1
Pytorch 1.9.0+cu111
Pytorch-lightning 1.7.4
Transformers 4.10.0

Install

To easily reproduce our results, you can install the environments by

pip install -r requirements.txt

Prepare IEMOCAP

Obtain IEMOCAP from USC

cd Dataset/IEMOCAP &&
python make_16k.py IEMOCAP_DIR &&
python gen_meta_label.py IEMOCAP_DIR &&
python generate_labels_sessionwise.py &&
cd ../..

Usage

Step One

Task Adaptation Pre-training process on IEMOCAP

bash bin/run_exp_iemocap_baseline.sh Dataset/IEMOCAP/Audio_16k/Dataset/IEMOCAP/labels_sess/label_2.json output_iemocap_train 2 TAPT 1

Step Two

Calculate how many clusters in the dataset

CUDA_VISIBLE_DEVICES=2 python clustering.py --datadir Dataset/IEMOCAP/Audio_16k/ --labeldir Dataset/IEMOCAP/labels_original_sess/label_2.json --save_name iemocap_train

If you face the problem ImportError: cannot import name 'Wav2Vec2ForXVector' from 'transformers' You could update your transformers version to 4.22.1

Step Three

Run with Random as initialization for Active Learning method

bash bin/TAPT_full_train.sh Dataset/IEMOCAP/Audio_16k/ Dataset/IEMOCAP/labels_sess_new/label_2.json output_full_1 3 TAPT 1 random output_iemocap_train_1/last.ckpt Least_confidence

Run with Clustering as initialization for Active Learning method

bash bin/TAPT_full_train.sh Dataset/IEMOCAP/Audio_16k/ Dataset/IEMOCAP/labels_sess_new/label_2.json output_full_1 3 TAPT 1 clustering output_iemocap_train_1/last.ckpt Entropy

Notions

The different Active Learning methods are listed here

Entropy
Least_confidence
alps
margin_confidence
bald
badge

License

MIT LICENSE

Clearloveyuan/AFTER