Evaluation of the Speech Resynthesis Capabilities of the VoicePrivacy Challenge Baseline B1
Ünal Ege Gaznepoğlu and Nils Peters
Presented as selected papers from the 2023 ISCA SPSC Symposium on Security and Privacy in Speech Communication, August 19, 2023. The 2023 ISCA SPSC Symposium is a satellite event of Interspeech 2023.
Abstract
Speaker anonymization systems continue to improve their ability to obfuscate the original speaker characteristics in a speech signal, but often create processing artifacts and unnatural sounding voices as a tradeoff. Many of those systems stem from the VoicePrivacy Challenge Baseline B1, using a neural vocoder to synthesize speech from an F0, x-vectors and bottleneck features-based speech representation. Inspired by this, we investigate the reproduction capabilities of the aforementioned baseline, to assess how successful the shared methodology is in synthesizing human-like speech. We use four objective metrics to measure speech quality, waveform similarity, and F0 similarity. Our findings indicate that both the speech representation and the vocoder introduces artifacts, causing an unnatural perception. A MUSHRA-like listening test on 18 subjects corroborate our findings, motivating further research on the analysis and synthesis components of the VoicePrivacy Challenge Baseline B1.
Listening test sound stimuli
8455-210777-0034.wav
p229_016_mic2.wav
p226_080_mic2.wav
p232_015_mic2.wav
5142-36377-0023.wav
p287_015_mic2.wav
1995-1837-0004.wav
8463-294828-0006.wav
6829-68771-0013.wav
p269_015_mic2.wav
7729-102255-0031.wav
8230-279154-0024.wav
p225_024_mic2.wav
260-123286-0018.wav
Paper (click to enlarge)
@techreport{2023_gaznepoglu_evaluation_vpc,
author = {Ünal Ege Gaznepoğlu, Nils Peters},
institution = {3rd ISCA Symp. on Security and Privacy in Speech Communication (SPSC)},
title = {Evaluation of the Speech Resynthesis Capabilities of the VoicePrivacy Challenge Baseline B1},
keywords = {},
year = {2023}}