ISCA Archive SMM 2022
ISCA Archive SMM 2022

Multi-task learning from Unlabelled Data to Improve Cross Language Speech Emotion Recognition

Hafiz Shehbaz Ali, Siddique Latif

Despite the recent progress in deep learning-based speech emotion recognition (SER), the performance of state-of-the-art systems significantly decreases in cross-language settings. The main reason is the lack of generalisation in SER systems due to the unavailability of larger training emotional labelled data in different languages. In this work, we present a novel multi-task learning (MTL) approach to effectively utilise unlabelled data to improve the generalisation as well as the performance of crosslanguage SER systems. In particular, we propose to use language and domain identification as auxiliary tasks, which facilities the proposed framework to learn from abundantly available language identification data. We evaluate the proposed model on publicly available datasets in four languages and achieve state-of-the-art performance.


doi: 10.21437/SMM.2022-4

Cite as: Ali, H.S., Latif, S. (2022) Multi-task learning from Unlabelled Data to Improve Cross Language Speech Emotion Recognition. Proc. Workshop on Speech, Music and Mind, 16-20, doi: 10.21437/SMM.2022-4

@inproceedings{ali22_smm,
  author={Hafiz Shehbaz Ali and Siddique Latif},
  title={{Multi-task learning from Unlabelled Data to Improve Cross Language Speech Emotion Recognition}},
  year=2022,
  booktitle={Proc. Workshop on Speech, Music and Mind},
  pages={16--20},
  doi={10.21437/SMM.2022-4}
}