Revise, Reason, and Recognize: LLM-Based Emotion Recognition via Emotion-Specific Prompts and ASR Error Correction

TL;DR

The paper investigates LLM-based emotion recognition from spoken language using emotion-specific prompts that fuse acoustic, linguistic, and psychological knowledge. It introduces the Revise-Reason-Recognize (R3) pipeline with ASR error correction to robustly handle imperfect transcripts and evaluates how different LLM training schemes (context-aware, in-context, instruction tuning) affect performance. Empirical results show that emotion-specific prompts, especially those leveraging linguistic triggers and ASR-emotion relations, combined with AEC and reasoning in R3, improve recognition, though ASR errors still impact performance. The work highlights the importance of prompt design and training strategies for robust, real-world emotion recognition from ASR-transcribed speech and offers guidance for deploying LLM-based emotion systems in noisy settings.

Abstract

Annotating and recognizing speech emotion using prompt engineering has recently emerged with the advancement of Large Language Models (LLMs), yet its efficacy and reliability remain questionable. In this paper, we conduct a systematic study on this topic, beginning with the proposal of novel prompts that incorporate emotion-specific knowledge from acoustics, linguistics, and psychology. Subsequently, we examine the effectiveness of LLM-based prompting on Automatic Speech Recognition (ASR) transcription, contrasting it with ground-truth transcription. Furthermore, we propose a Revise-Reason-Recognize prompting pipeline for robust LLM-based emotion recognition from spoken language with ASR errors. Additionally, experiments on context-aware learning, in-context learning, and instruction tuning are performed to examine the usefulness of LLM training schemes in this direction. Finally, we investigate the sensitivity of LLMs to minor prompt variations. Experimental results demonstrate the efficacy of the emotion-specific prompts, ASR error correction, and LLM training schemes for LLM-based emotion recognition. Our study aims to refine the use of LLMs in emotion recognition and related domains.

Figures (2)

• Figure 1: Emotion-specific prompts used in this work.
• Figure 2: Examples of LLM reasoning with emotion-specific knowledge.