AI and speaker rights

By using special new technologies, it is now possible to generate human-like voices with such remarkable precision that they sound almost identical to the original human voice being synthesised. However, if the AI system is trained using human voice recordings, this gives rise to a number of legal issues that need to be given careful consideration.

In principle, it is against the law to train an AI system with human voices without first obtaining the consent of the data subject. This applies in particular if the voices contain identifiable information or exhibit specific individual characteristics (e.g. where an individual has a distinctive dialect or his/her voice is highly recognisable). The potential legal consequences of such unlawful use can be found in the law governing personal rights, in data protection legislation and possibly copyright law.

Below is a brief list of the relevant legal issues:

• Personal rights: Only the person concerned has the right to his/her own voice, and it is not transferable, even if it is a synthesised version of that person's own voice. Failure to respect these rights can result in (among other things) claims for damages and actions for an injunction.

• Data protection: A person's voice is classified as biometric data and enjoys special protection under Article 9 GDPR. Where synthesis is misused, there are unlikely to be any grounds for justification within the meaning of Article 9(2), which may also result in claims for damages.

• Copyright: As a rule, works (of a linguistic nature) generated by AI do not establish a copyright because the AI itself does not engage in creative activity. In addition, the Austrian Copyright Act requires what is referred to as an "act of creation", i.e. a human thought process, for a work to enjoy copyright protection. This raises the question of whether the voice itself, which is used for training purposes, enjoys copyright protection. However, as a human voice alone lacks the required level of creativity, it is not protected by copyright in its natural form. It is a different matter, however, if the voice is taken from a sound recording, for instance. If a work of sound art within the meaning of the Copyright Act exists, then (unauthorised) use of (parts of) this sound recording may constitute a copyright infringement. In such a case, the copyright holder may assert various claims, e.g. for injunctive relief, removal of the work, payment of an appropriate fee, damages and publication of a judgment (in the case of fault), possibly combined with an application for an interim injunction.