Can we steal your vocal identity from the Internet?: Initial investigation of cloning Obama’s voice using GAN, WaveNet and low-quality found data

Abstract

Thanks to the growing availability of spoofing databases and rapid advances in using them, systems for detecting voice spoofing attacks are becoming more and more capable, and error rates close to zero are being reached for the ASVspoof2015 database. However, speech synthesis and voice conversion paradigms that are not considered in the ASVspoof2015 database are appearing.Such examples include direct waveform modelling and generative adversarial networks. We also need to investigate the feasibility of training spoofing systems using only low-quality found data. For that purpose, we developed a generative adversarial networkbased speech enhancement system that improves the quality of speech data found in publicly available sources. Using the enhanced data, we trained state-of-the-art text-to-speech and voice conversion models and evaluated them in terms of perceptual speech quality and speaker similarity. The results show that the enhancement models significantly improved the SNR of low-quality degraded data found in publicly available sources and that they significantly improved the perceptual cleanliness of the source speech without significantly degrading the naturalness of the voice. However, the results also show limitations when generating speech with the low-quality found data.

Voice Mimicry Attacks Assisted by Automatic Speaker Verification

“…Other risks could include fabricating a 'digital clone' of someone using machine learning -recent warning examples are provided by the so-called deepfakes [26,27,28], realistic-appearing but fabricated or tampered videos portraying a targeted person created with the aid of deep learning (the interested reader is pointed to to [29] for a detailed review of potential societal, ethical and legal implications of deepfakes). In the context of speaker verification in specific, [30] addressed voice cloning of a well-known celebrity (the former US president Barack Obama). Even if the result was essentially negative (the cloned voice samples were detectable as artificial ones using a spoofing countermeasure), machine learning, including voice cloning techniques, do not stand still.…”

End-to-End Text-to-Speech for Low-Resource Languages by Cross-Lingual Transfer Learning

“…Recent research on end-to-end text-to-speech (TTS) [1,2,3,4,5,6] has gained success in terms of human-like and highquality generated speech. Moreover, with regard to cloning prosody style or speaker characteristics, end-to-end TTS systems also demonstrate a powerful capability [7,8,9,10,11]. However, training end-to-end TTS systems requires large quantities of text-audio paired data.…”

Can We Use Speaker Recognition Technology to Attack Itself? Enhancing Mimicry Attacks Using Automatic Target Speaker Selection

“…In line with the recent EU's General Data Protection Regulation (GDPR), intended to protect the privacy of its citizens, it is important to assess risks associated with multimedia data in the public domain. A recent study [17] has attempted voice cloning of celebrity voices based on found data using a pre-defined target speaker. The cloned voice samples were, however, detectable as spoofed speech.…”