Synthetic Speech Attribution via Prototypical Networks

2026-09-30

AI Summary

This study addresses the lack of interpretability in black-box models for synthetic speech attribution by proposing, for the first time, a prototype-based interpretable recognition framework. The method employs the ProtoPNet architecture to extract spectrogram features and identifies generative systems through comparative reasoning against representative prototype samples, thereby providing example-level decision justifications. Experimental evaluations on the MLAAD dataset demonstrate that this framework consistently outperforms existing baseline methods across closed-set, cross-lingual, and open-set scenarios. Ultimately, the proposed approach effectively unifies high classification accuracy with strong interpretability, offering a transparent alternative to conventional opaque models in synthetic speech forensics.

Abstract

Synthetic speech attribution aims to identify the generative system responsible for a speech signal, but current approaches typically rely on black-box neural networks that provide limited insight into their decisions. This work investigates prototype-based networks as an interpretable alternative, where predictions are grounded in comparisons with representative training examples. We adapt ProtoPNet to spectrogram-based speech representations and evaluate the proposed framework on the MLAAD dataset under closed-set, cross-lingual, and open-set conditions. Experiments show that prototype-based reasoning achieves competitive or improved attribution performance compared with the baseline while enabling example-based explanations. These results highlight that interpretability and performance can be jointly achieved in synthetic speech attribution through prototype-based modeling.

Problem

• Synthetic Speech Attribution
• Interpretability
• Black-box Neural Networks
• Speech Forensics

Innovation

• Synthetic Speech Attribution
• Prototypical Networks
• ProtoPNet
• Interpretability
• Spectrogram