- The model is based on this colab notebook https://colab.research.google.com/github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb#scrollTo=0Tl6iKAUR4EL
- I used a python 3.9 environment.
- Install all requirements via pip install -r requirements.txt.
- When choosing a wav2vec2 model, make sure it is a base model trained on self attention. You can read about some models I tested in the train.py.
- Download dataset from https://www.openslr.org/resources/99/NonverbalVocalization.tgz and extract
- Execute python train.py. Training data will be preprocessed within this script.
- Execute python test_local.py to load a model from the local folder myrun2
- Execute python test_online.py to load my pretrained model from huggingface hub.
- To use wav2vec2 for sequence classification might also be a good approach.