Voice Conversion Improves Cross-Domain Robustness for Spoken Arabic Dialect Identification

2025-05-30

AI Summary

Arabic Dialect Identification (ADI) suffers from poor cross-domain generalization, primarily due to speaker bias and domain shift. To address this, we propose the first speaker-decoupled ADI framework leveraging voice conversion (VC), jointly optimizing end-to-end dialect classification and cross-domain robust training. Our method explicitly disentangles speaker identity from dialect-relevant features via VC-based representation learning, thereby mitigating speaker bias and enhancing generalization to unseen recording domains. We further introduce the first real-world multi-domain ADI benchmark—comprising four newly collected, naturally diverse domains—to rigorously evaluate cross-domain robustness. Extensive experiments demonstrate that our approach achieves up to a 34.1% absolute accuracy improvement over prior methods in cross-domain evaluation, establishing new state-of-the-art performance. All code, models, and datasets are publicly released.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Conversational AI/Dialog Systems
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Arabic dialect identification (ADI) systems are essential for large-scale data collection pipelines that enable the development of inclusive speech technologies for Arabic language varieties. However, the reliability of current ADI systems is limited by poor generalization to out-of-domain speech. In this paper, we present an effective approach based on voice conversion for training ADI models that achieves state-of-the-art performance and significantly improves robustness in cross-domain scenarios. Evaluated on a newly collected real-world test set spanning four different domains, our approach yields consistent improvements of up to +34.1% in accuracy across domains. Furthermore, we present an analysis of our approach and demonstrate that voice conversion helps mitigate the speaker bias in the ADI dataset. We release our robust ADI model and cross-domain evaluation dataset to support the development of inclusive speech technologies for Arabic.

Problem

• Improves cross-domain robustness for Arabic dialect identification
• Addresses poor generalization of current ADI systems
• Mitigates speaker bias in ADI datasets using voice conversion

Innovation

• Voice conversion enhances cross-domain ADI robustness
• State-of-the-art performance achieved with voice conversion
• Voice conversion reduces speaker bias in ADI