Speaker Anonymization Using X-vector and Neural Waveform Models

Abstract

The social media revolution has produced a plethora of web services to which users can easily upload and share multimedia documents. Despite the popularity and convenience of such services, the sharing of such inherently personal data, including speech data, raises obvious security and privacy concerns. In particular, a user's speech data may be acquired and used with speech synthesis systems to produce high-quality speech utterances which reflect the same user's speaker identity. These utterances may then be used to attack speaker verification systems. One solution to mitigate these concerns involves the concealing of speaker identities before the sharing of speech data. For this purpose, we present a new approach to speaker anonymization. The idea is to extract linguistic and speaker identity features from an utterance and then to use these with neural acoustic and waveform models to synthesize anonymized speech. The original speaker identity, in the form of timbre, is suppressed and replaced with that of an anonymous pseudo identity. The approach exploits state-of-the-art x-vector speaker representations. These are used to derive anonymized pseudo speaker identities through the combination of multiple, random speaker x-vectors. Experimental results show that the proposed approach is effective in concealing speaker identities. It increases the equal error rate of a speaker verification system while maintaining high quality, anonymized speech.

Preventing Fake Information Generation Against Media Clone Attacks

“…Speech data contain much sensitive information that can be used to identify the speaker. In this subsection, we introduce our initial work on speaker anonymization [51]. The speaker anonymization technique we have developed can conceal a speaker's identity while keeping other factors such as linguistic information, naturalness, and quality unchanged.…”

A Comparative Study of Speech Anonymization Metrics

“…Finally, the primary baseline of the VoicePrivacy Challenge 2020 (VPC) [25] uses a neural synthesizer [9,26] to synthesize speech given the target x-vector and fundamental frequency and bottleneck features extracted from the source.…”

“…Their purpose is to transform speech signals in order to preserve all content except features related with the speaker identity. These techniques include noise addition [4], speech transformation [5], voice conversion [6,7,8], speech synthesis [9], or adversarial learning [10]. As a privacy preservation mechanism, they must achieve a suitable privacy/utility trade-off.…”

Personal data protection and academia: GDPR issues and multi-modal data-collections "in the wild"

“…Advanced voice-based speaker anonymization aims to suppress the speaker identity and is mostly based on voice transformation techniques, changing source, or filter characteristics of the speech (Pobar & Ipšić, 2014). Recent research suggested to use x-vector speaker representations to suppress the timbre of a speaker, and thus hindering the speaker identification (Fang et al, 2019). Although these techniques can preserve the textual context and may sound natural, they cannot sustain the emotional expression of a speaker, as they have never aimed that.…”