SER Wav2vec2 Large Xlsr 53 Eng Zho All Age

🚀 Cross-Lingual Cross-Age Group Adaptation for Low-Resource Elderly Speech Emotion Recognition

This project fine-tunes a model on English and Chinese data from all - age speakers for elderly speech emotion recognition, aiming to address low - resource scenarios.

🚀 Quick Start

This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 on English and Chinese data from all - age speakers. It is trained on the training sets of multiple datasets, including [CREMA - D](https://github.com/CheyneyComputerScience/CREMA - D), [CSED](https://github.com/AkishinoShiame/Chinese - Speech - Emotion - Datasets), ElderReact, [ESD](https://github.com/HLTSingapore/Emotional - Speech - Data), IEMOCAP, and [TESS](https://www.kaggle.com/datasets/ejlok1/toronto - emotional - speech - set - tess).

When using this model, ensure that your speech input is sampled at 16kHz.

The scripts for training and evaluation can be found at https://github.com/HLTCHKUST/elderly_ser/tree/main.

📚 Documentation

Evaluation Results

For detailed information such as the statistics of train, valid, and test data, please refer to our paper on arXiv. The paper also presents the model's speech emotion recognition performances on: English - All, Chinese - All, English - Elderly, Chinese - Elderly, English - Adults, Chinese - Adults.

Citation

Our paper will be published at INTERSPEECH 2023. You can currently access it on arXiv. If you find our work useful, please cite our paper as follows:

@misc{cahyawijaya2023crosslingual,
      title={Cross-Lingual Cross-Age Group Adaptation for Low-Resource Elderly Speech Emotion Recognition}, 
      author={Samuel Cahyawijaya and Holy Lovenia and Willy Chung and Rita Frieske and Zihan Liu and Pascale Fung},
      year={2023},
      eprint={2306.14517},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

📄 License

This project is licensed under CC - BY - SA 4.0.

📦 Additional Information