Vicomtech develops biometric security and anti-spoofing technology based on Artificial Intelligence to prevent identity theft by voice

The great advances in Artificial Intelligence (AI) in different fields such as the processing of speech and natural language are allowing, among other developments, to synthesize or clone a voice with great realism, to the point of making it difficult to distinguish a real voice from another artificially generated. This ability to clone voices, even of deceased people, opens a wide range of possibilities for innovative narrative techniques, for the customization of conversational assistants or even for dubbing, among others.

Even so, voice cloning and deep fake also invite the development of malicious techniques mainly related to phishing and all its possible applications.

At Vicomtech we develop technology based on Artificial Intelligence and deep neural networks in order to detect whether a voice is real or has been generated using voice synthesis or voice conversion technologies. Our technologies even allow us to determine if a voice corresponds to a recording made covertly to a legitimate user in order to authenticate in a voice verification system and to find out phishing cases.

This voice anti-spoofing technology can be applied in many domains and environments, such as in audiovisual production, voice access controls, or in telephone call centers.

At Vicomtech we have a long history of research into these technologies, working with data applicable to any of these fields. In fact, we have recently generated a new database called ASVspoophone, based on the audios of the original ASVspoof2019 database, built for the international anti-spoofing challenge ASVspoof2019. The new ASVspoophone database includes the audios transferred by the telephone channel and Vicomtech has already generated the first base model capable of discerning between real, recorded, or generated voices using synthesis and voice conversion techniques. These results have been forwarded to the international Text, Speech, and Dialogue 2020 conference.

In addition to voice anti-spoofing techniques, various research groups at the Center are working together on the development and integration of natural language processing technology, artificial vision and network information with the aim of responding to the challenges that arise. related to the detection of hoaxes and fake news in different sectors of activity.