mlsanityData to 13 Oct 2026

HARNESS: Lightweight Distilled Arabic Speech Foundation Models

arXiv 2604.14186 ↗First posted 31 Mar 2026eess.AS cs.AI cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

Large self-supervised speech (SSL) models achieve strong downstream performance, but their size limits deployment in resource-constrained settings. We present HArnESS, an Arabic-centric self-supervised speech model family trained from scratch with iterative self-distillation, together with lightweight student variants that offer strong accuracy-efficiency trade-offs on Automatic Speech Recognition (ASR), Dialect Identification (DID), and Speech Emotion Recognition (SER). Our approach begins with a large bilingual Arabic-English teacher and progressively distills its knowledge into compressed student models while preserving Arabic-relevant acoustic and paralinguistic representations. We further study PCA-based compression of the teacher supervision signal to better match the capacity of shallow and thin students. Compared with HuBERT and XLS-R, HArnESS consistently improves performance on Arabic downstream tasks, while the compressed models remain competitive under substantial structural reduction. These results position HArnESS as a practical and accessible Arabic-centric SSL foundation for real-world speech applications.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Builds on 24 of its 35 references that are in the map. First posted on arXiv 31 Mar 2026.

Builds on

24 of 35 references.

References · 24

  1. 2020wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representationscited by 3,870
  2. 2021HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Unitscited by 2,313
  3. 2021WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processingcited by 1,625
  4. 2019fairseq: A Fast, Extensible Toolkit for Sequence Modelingcited by 1,487
  5. 2019Common Voice: A Massively-Multilingual Speech Corpuscited by 1,366
  6. 2022data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Languagecited by 721
  7. 2021SUPERB: Speech processing Universal PERformance Benchmarkcited by 657
  8. 2021XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scalecited by 594
  9. 2021GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10,000 Hours of Transcribed Audiocited by 393
  10. 2021W2v-BERT: Combining Contrastive Learning and Masked Language Modeling for Self-Supervised Speech Pre-Trainingcited by 342
  11. 2022Self-Supervised Speech Representation Learning: A Reviewcited by 258
  12. 2021DistilHuBERT: Speech Representation Learning by Layer-wise Distillation of Hidden-unit BERTcited by 130
12 more
  1. 2023ML-SUPERB: Multilingual Speech Universal PERformance Benchmarkcited by 70
  2. 2025Fanar: An Arabic-Centric Multimodal Generative AI Platformcited by 67
  3. 2017Speech Recognition Challenge in the Wild: Arabic MGB-3cited by 60
  4. 2022LightHuBERT: Lightweight and Configurable Speech Representation Learning with Once-for-All Hidden-Unit BERTcited by 45
  5. 2021QASR: QCRI Aljazeera Speech Resource -- A Large Scale Annotated Arabic Speech Corpuscited by 38
  6. 2023DPHuBERT: Joint Distillation and Pruning of Self-Supervised Speech Modelscited by 33
  7. 2022Deep versus Wide: An Analysis of Student Architectures for Task-Agnostic Knowledge Distillation of Self-Supervised Speech Modelscited by 28
  8. 2022FitHuBERT: Going Thinner and Deeper for Knowledge Distillation of Speech Self-Supervised Learningcited by 24
  9. 2016The MGB-2 Challenge: Arabic Multi-Dialect Broadcast Media Recognitioncited by 22
  10. 2023Yet Another Model for Arabic Dialect Identificationcited by 7
  11. 2024SKILL: Similarity-aware Knowledge distILLation for Speech Self-Supervised Learningcited by 5
  12. 2025ArabEmoNet: A Lightweight Hybrid 2D CNN-BiLSTM Model with Attention for Robust Arabic Speech Emotion Recognitioncited by 2

Built on by

Not cited yet.