New AMRD method enables lightweight speech emotion recognition models

Researchers have developed Adaptive Multi-teacher Relational Distillation (AMRD), a novel method to create lightweight speech emotion recognition (SER) models suitable for on-device applications. AMRD addresses challenges in knowledge distillation by adaptively weighting teachers based on their batch reliability and incorporating a relational distillation loss to capture inter-sample structural information. Experiments on the IEMOCAP and CREMA-D datasets demonstrated that AMRD surpasses traditional single-teacher distillation baselines across various student architectures. AI

IMPACT Enables more efficient on-device speech emotion recognition by creating lightweight models.

RANK_REASON The cluster describes a new research paper detailing a novel method for speech emotion recognition. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• Adaptive Multi-teacher Relational Distillation
• CREMA-D
• IEMOCAP: interactive emotional dyadic motion capture database
• One-Class SVMs Challenges in Audio Detection and Classification Applications
• Speech Emotion Recognition Using Machine Learning

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →