Neural Audio Codec for Robust Audio Deepfake Detection

2026-09-30

AI Summary

Low-bitrate audio coding significantly degrades deepfake detection performance. To address this issue, this work proposes a forensics-preserving neural audio codec that establishes "forensic transparency" as an explicit design objective. By employing detector-guided fine-tuning of pretrained models, the proposed approach optimizes detection robustness while preserving the original quantization pipeline and bitrate constraints. Notably, this method achieves improved generalization across multiple detectors under a single supervision signal. Experimental results on the ASVspoof 2019 LA dataset demonstrate that the proposed codec reduces the equal error rate (EER) by 49.8 percentage points compared to the original DAC baseline, while maintaining comparable reconstruction quality.

Abstract

Audio deepfake detectors are typically evaluated on uncompressed audio, although real-world audio often undergoes low-bitrate coding. In this work, we investigate how audio coding affects deepfake detection across codecs, bitrates, and detectors, finding higher errors at lower rates. A mixed-pair protocol isolates codec-induced changes in bona fide and spoof audio, revealing asymmetric, codec-dependent failures: low-rate DAC and EnCodec mainly degrade bona fide detection, whereas X-Codec shows a stronger spoof-side limitation. Motivated by these, we propose a forensic-preserving neural audio codec (FP-NAC), which fine-tunes a pretrained codec using a detector-guided objective while preserving its native hard quantization path and bitrate. On ASVspoof 2019 LA, FP-NAC reduces EER by up to 49.8~pp compared with the original DAC at 0.5~kbps while maintaining comparable reconstruction quality. Although supervised by only one detector, FP-NAC improves performance across multiple detectors, highlighting forensic transparency as a codec design objective alongside perceptual quality. Our codes are available at https://github.com/kjungwoo03/FP-NAC.

Problem

• audio deepfake detection
• neural audio codec
• low-bitrate coding
• forensic preservation

Innovation

• Forensic-preserving neural audio codec
• Audio deepfake detection
• Low-bitrate coding
• Detector-guided objective
• Forensic transparency