Distinguishing Neural Speech Synthesis Models Through Fingerprints in Speech Waveforms

TL;DR

This paper addresses attribution of neural speech synthesis sources by detecting model-specific fingerprints in TTS waveforms, focusing on both acoustic models and vocoders. It introduces a fingerprint extraction pipeline using LFCC features and a Res2Net-based classifier trained on LibriTTS, and analyzes how fingerprints from each component manifest and interact. The key findings show that both vocoders and acoustic models leave distinct fingerprints, with vocoder fingerprints being more dominant and capable of masking acoustic-model fingerprints, suggesting practical utility for source attribution and forensics. The work highlights implications for designing robust attribution systems and outlines future directions, including end-to-end TTS and cross-language considerations.

Abstract

Recent strides in neural speech synthesis technologies, while enjoying widespread applications, have nonetheless introduced a series of challenges, spurring interest in the defence against the threat of misuse and abuse. Notably, source attribution of synthesized speech has value in forensics and intellectual property protection, but prior work in this area has certain limitations in scope. To address the gaps, we present our findings concerning the identification of the sources of synthesized speech in this paper. We investigate the existence of speech synthesis model fingerprints in the generated speech waveforms, with a focus on the acoustic model and the vocoder, and study the influence of each component on the fingerprint in the overall speech waveforms. Our research, conducted using the multi-speaker LibriTTS dataset, demonstrates two key insights: (1) vocoders and acoustic models impart distinct, model-specific fingerprints on the waveforms they generate, and (2) vocoder fingerprints are the more dominant of the two, and may mask the fingerprints from the acoustic model. These findings strongly suggest the existence of model-specific fingerprints for both the acoustic model and the vocoder, highlighting their potential utility in source identification applications.

Figures (4)

• Figure 1: TTS pipeline and t-SNE projection of acoustic model and vocoder fingerprints.
• Figure 2: Fingerprint extraction pipeline.
• Figure 3: Confusion matrices of experiments in our study.
• Figure 4: t-SNE representation of fingerprints extracted in our experiments.