AMRD: Adaptive Multi-Teacher Relational Distillation for Lightweight Speech Emotion Recognition

Hugging Face Daily Papers Papers

Summary

This paper introduces AMRD, an adaptive multi-teacher relational distillation method for compressing large self-supervised speech emotion recognition models into lightweight student models for edge devices. It addresses teacher reliability variation and relational structure loss, showing improvements on IEMOCAP and CREMA-D datasets.

On-device speech emotion recognition (SER) is critical for real-time applications, yet large self-supervised models that excel at SER are too costly for edge devices. Multi-teacher knowledge distillation can compress them into a lightweight student, but two challenges remain: teacher reliability varies across batches, and logit-level distillation ignores inter-sample relational structure. We propose Adaptive Multi-teacher Relational Distillation (AMRD) to address both. A one-class SVM on each teacher's logit similarity matrix assigns per-batch weights favoring more coherent teachers. A relational distillation loss aligns teacher and student similarity matrices, capturing structure that logit matching misses. On IEMOCAP and CREMA-D datasets across four student architectures, AMRD outperforms single-teacher distillation baselines in most settings, and ablations confirm both components yield complementary gains.
Original Article
View Cached Full Text

Cached at: 07/31/26, 05:52 AM

Paper page - AMRD: Adaptive Multi-Teacher Relational Distillation for Lightweight Speech Emotion Recognition

Source: https://huggingface.co/papers/2607.25289

Abstract

On-devicespeechemotionrecognition(SER)iscriticalforreal-timeapplications,yetlargeself-supervisedmodelsthatexcelatSERaretoocostlyforedgedevices.Multi-teacherknowledgedistillationcancompressthemintoalightweightstudent,buttwochallengesremain:teacherreliabilityvariesacrossbatches,andlogit-leveldistillationignoresinter-samplerelationalstructure.WeproposeAdaptiveMulti-teacherRelationalDistillation(AMRD)toaddressboth.Aone-classSVMoneachteacher’slogitsimilaritymatrixassignsper-batchweightsfavoringmorecoherentteachers.Arelationaldistillationlossalignsteacherandstudentsimilaritymatrices,capturingstructurethatlogitmatchingmisses.OnIEMOCAPandCREMA-Ddatasetsacrossfourstudentarchitectures,AMRDoutperformssingle-teacherdistillationbaselinesinmostsettings,andablationsconfirmbothcomponentsyieldcomplementarygains.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.25289

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.25289 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.25289 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.25289 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Do We Really Need Multimodal Emotion Language Models Larger Than 1B Parameters?

arXiv cs.AI

The paper proposes Light-MER, a lightweight multimodal emotion recognition framework that uses knowledge distillation from an 8B teacher model to a sub-1B student, achieving state-of-the-art performance with significantly higher inference efficiency, challenging the necessity of models larger than 1B parameters.