The CONILIUM proposition for Odyssey Emotion Challenge Leveraging major class with complex annotations - LIUM - Equipe Language and Speech Technology
Conference Papers Year : 2024

The CONILIUM proposition for Odyssey Emotion Challenge Leveraging major class with complex annotations

Abstract

This paper describes the contribution of the CONILIUM team in the Odyssey Emotion Recognition Challenge. Our system focuses on predicting categorical emotions from speech recordings in the MSP-Podcast corpus. Focusing on the training protocol, we investigated several approaches to improve emotion recognition accuracy. Different pre-trained models (WavLM-large, Wav2vec2-large, Hubert-large) were evaluated as feature extractors. An agreement-aware loss functions based on all secondary annotations is proposed that consider the disagreement among annotators and the ambiguity of emotional labeling during training. An idea of keeping only samples with high agreement annotation in the training process shows the benefit of using all annotations by all annotators. Our best performing system utilized WavLM-large as the upstream model, weighted binary cross entropy with secondary labels as the loss function, and a post-processing step that adjusted the decision threshold. This model achieved an F1-Macro score of 0.361 on the development set, 0.335 on the test set, which is a significant improvement compare to the provided baseline. We also explore characteristics of Easy and Difficult samples based on their prediction performance consistency across different models.
Fichier principal
Vignette du fichier
MSP_Podcast_Challenge_Odyssey_2024_.pdf (332.17 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04600047 , version 1 (04-06-2024)

Identifiers

Cite

Meysam Shamsi, Lara Gauder, Marie Tahon. The CONILIUM proposition for Odyssey Emotion Challenge Leveraging major class with complex annotations. The Speaker and Language Recognition Workshop (Odyssey), Jun 2024, Quebec, Canada. pp.281-287, ⟨10.21437/odyssey.2024-40⟩. ⟨hal-04600047⟩
46 View
44 Download

Altmetric

Share

More