SEA-Spoof: Bridging The Gap in Multilingual Audio Deepfake Detection for South-East Asian

Authors: Jinyang Wu, Nana Hou, Zihan Pan, Qiquan Zhang, Sailor Hardik Bhupendra, Soumik Mondal

Published: 2025-09-24 08:11:51+00:00

Comment: 5 pages, 1 figure, 3 tables

AI Summary

This paper introduces SEA-Spoof, the first large-scale Audio Deepfake Detection (ADD) dataset specifically designed for South-East Asian (SEA) languages, addressing the critical gap where current models fail due to data scarcity and linguistic mismatches. Spanning over 300 hours across six SEA languages, SEA-Spoof includes paired real and spoof speech generated by diverse state-of-the-art systems. Benchmarking reveals severe cross-lingual performance degradation, which is significantly mitigated by fine-tuning models on SEA-Spoof, thereby highlighting its importance for robust, cross-lingual fraud detection.

Abstract

The rapid growth of the digital economy in South-East Asia (SEA) has amplified the risks of audio deepfakes, yet current datasets cover SEA languages only sparsely, leaving models poorly equipped to handle this critical region. This omission is critical: detection models trained on high-resource languages collapse when applied to SEA, due to mismatches in synthesis quality, language-specific characteristics, and data scarcity. To close this gap, we present SEA-Spoof, the first large-scale Audio Deepfake Detection (ADD) dataset especially for SEA languages. SEA-Spoof spans 300+ hours of paired real and spoof speech across Tamil, Hindi, Thai, Indonesian, Malay, and Vietnamese. Spoof samples are generated from a diverse mix of state-of-the-art open-source and commercial systems, capturing wide variability in style and fidelity. Benchmarking state-of-the-art detection models reveals severe cross-lingual degradation, but fine-tuning on SEA-Spoof dramatically restores performance across languages and synthesis sources. These results highlight the urgent need for SEA-focused research and establish SEA-Spoof as a foundation for developing robust, cross-lingual, and fraud-resilient detection systems.

Key findings

State-of-the-art deepfake detection models, while effective on benchmarks like ASVspoof5, suffer severe performance degradation (e.g., MoLEx dropping from 1.25% to 43.8% EER) when applied to SEA-Spoof languages. Fine-tuning on SEA-Spoof dramatically restores detection accuracy (e.g., MoLExft achieving 0.2% EER), demonstrating the dataset's value. Commercial spoofing models generally generate harder-to-detect fakes, and detection difficulty varies significantly across different SEA languages, with Tamil and Malay proving most challenging.

Approach

The authors developed SEA-Spoof, a large-scale dataset of paired real and spoofed speech across six SEA languages (Tamil, Hindi, Thai, Indonesian, Malay, Vietnamese), generated using 10 open-source and 4 commercial TTS/VC systems. They benchmarked state-of-the-art audio deepfake detection models on this dataset, observing significant performance degradation, and then demonstrated that fine-tuning these models on SEA-Spoof dramatically improves their robustness for SEA languages.

Datasets

SEA-Spoof, ASVspoof5, ASVspoof 2019 logical access (LA), MLAAD, DFADD, Fake-or-Real, Mozilla Common Voice, Indic Speech Corpora, GigaSpeech2, Malay Conversational Speech Corpus, Malaysian YouTube Whisper-Large set, Thai Dialect Corpus, VIVOS corpus

Model(s)

AASIST, AASIST3, MoLEx, WavLM (as SSL feature extractor for MoLEx)

Author countries

Singapore, Australia

← Previous