The Sound of Silence: Efficiency of First Digit Features in Synthetic Audio Detection

TL;DR

This paper explores how first digit statistics from MFCC coefficients can effectively detect synthetic speech, offering a lightweight and robust method that achieves over 90% accuracy across various forgery techniques.

Contribution

It introduces a novel, computationally-efficient detection method based on first digit analysis of MFCCs, improving robustness without complex neural architectures.

Findings

Achieves over 90% accuracy on ASVSpoof dataset

Effective across multiple synthetic speech algorithms

Lightweight and computationally efficient

Abstract

The recent integration of generative neural strategies and audio processing techniques have fostered the widespread of synthetic speech synthesis or transformation algorithms. This capability proves to be harmful in many legal and informative processes (news, biometric authentication, audio evidence in courts, etc.). Thus, the development of efficient detection algorithms is both crucial and challenging due to the heterogeneity of forgery techniques. This work investigates the discriminative role of silenced parts in synthetic speech detection and shows how first digit statistics extracted from MFCC coefficients can efficiently enable a robust detection. The proposed procedure is computationally-lightweight and effective on many different algorithms since it does not rely on large neural detection architecture and obtains an accuracy above 90\% in most of the classes of the ASVSpoof dataset.