Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Textualized and Feature-based Models for Compound Multimodal Emotion Recognition in the Wild

Nicolas Richet, Soufiane Belharbi, Haseeb Aslam, Meike Emilie Schadt, Manuela González-González, Gustave Cortal, Alessandro Lameiras Koerich, Marco Pedersoli, Alain Finkel, Simon Bacon, Eric Granger

TL;DR

Comparing the potential of text- and feature-based approaches for compound multimodal ER in videos indicates that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild.

Abstract

Systems for multimodal emotion recognition (ER) are commonly trained to extract features from different modalities (e.g., visual, audio, and textual) that are combined to predict individual basic emotions. However, compound emotions often occur in real-world scenarios, and the uncertainty of recognizing such complex emotions over diverse modalities is challenging for feature-based models. As an alternative, emerging large language models (LLMs) like BERT and LLaMA can rely on explicit non-verbal cues that may be translated from different non-textual modalities (e.g., audio and visual) into text. Textualization of modalities augments data with emotional cues to help the LLM encode the interconnections between all modalities in a shared text space. In such text-based models, prior knowledge of ER tasks is leveraged to textualize relevant non-verbal cues such as audio tone from vocal expressions, and action unit intensity from facial expressions. Since the pre-trained weights are publicly available for many LLMs, training on large-scale datasets is unnecessary, allowing to fine-tune for downstream tasks such as compound ER (CER). This paper compares the potential of text- and feature-based approaches for compound multimodal ER in videos. Experiments were conducted on the challenging C-EXPR-DB dataset in the wild for CER, and contrasted with results on the MELD dataset for basic ER. Our results indicate that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild. However, higher accuracy can be achieved when the video data has rich transcripts. Our code is available.

Textualized and Feature-based Models for Compound Multimodal Emotion Recognition in the Wild

TL;DR

Comparing the potential of text- and feature-based approaches for compound multimodal ER in videos indicates that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild.

Abstract

Systems for multimodal emotion recognition (ER) are commonly trained to extract features from different modalities (e.g., visual, audio, and textual) that are combined to predict individual basic emotions. However, compound emotions often occur in real-world scenarios, and the uncertainty of recognizing such complex emotions over diverse modalities is challenging for feature-based models. As an alternative, emerging large language models (LLMs) like BERT and LLaMA can rely on explicit non-verbal cues that may be translated from different non-textual modalities (e.g., audio and visual) into text. Textualization of modalities augments data with emotional cues to help the LLM encode the interconnections between all modalities in a shared text space. In such text-based models, prior knowledge of ER tasks is leveraged to textualize relevant non-verbal cues such as audio tone from vocal expressions, and action unit intensity from facial expressions. Since the pre-trained weights are publicly available for many LLMs, training on large-scale datasets is unnecessary, allowing to fine-tune for downstream tasks such as compound ER (CER). This paper compares the potential of text- and feature-based approaches for compound multimodal ER in videos. Experiments were conducted on the challenging C-EXPR-DB dataset in the wild for CER, and contrasted with results on the MELD dataset for basic ER. Our results indicate that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild. However, higher accuracy can be achieved when the video data has rich transcripts. Our code is available.
Paper Structure (10 sections, 1 equation, 3 figures, 11 tables)

This paper contains 10 sections, 1 equation, 3 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Feature-based Modeling
  3. Text-based Modeling
  4. Results and Discussion
  5. Experimental Methodology
  6. Results for Basic (MELD) and Compound (C-EXPR-DB$^*$) Emotions
  7. 7th ABAW CER Challenge Results
  8. Computation Time
  9. Challenges of Textualizing Modalities
  10. Conclusion

Figures (3)

  • Figure 1: Models for compound multimodal ER in videos. (a) In the feature-based approach, a dedicated feature extractor produces embeddings for each input modality. A feature-level fusion module then combines embeddings from all different modalities to produce joint feature representations for classification. (b) In the text-based approach, textual descriptions are extracted for nonverbal modalities, such as audio and visual. These texts are combined with verbal cues (i.e., text transcripts) and fed to an LLM as a joint textual representation for classification.
  • Figure 2: A common feature-based approach used for multimodal CER.
  • Figure 3: A common text-based approach used for multimodal CER, where non-verbal modalities are textualized.