In the multifaceted world of artificial intelligence (AI), a wide range of applications has emerged, from face and speech recognition to advanced speech synthesis. Particularly fascinating and topical is the development of technologies capable of synthesizing human voices with remarkable precision. Such AI systems can capture and recreate the nuances, pitches, and idiosyncrasies of human voices so well that they sound nearly identical to the original voices.
As a writer of nearly 1100 blogposts myself, I’ve been on a quest to find such technologies to transform my written content into dynamic, engaging video and audio formats. By extending my content into these new formats, I could expand my reach and make my content accessible in innovative ways.
However, in my research, I found that these technological advances also raise a number of legal issues that need to be carefully considered. A key question is whether an AI vendor can simply use voices of real people to train its AI.
Basically, the answer to this question is a clear no. Training an AI with human voices falls under the legal framework of personal rights and possibly copyright. AI providers may not simply use voices for their models without obtaining the explicit consent of the data subjects. This is especially true if the voices contain identifiable information or have specific individual characteristics. A provider who violates these provisions could face severe legal consequences.
Therefore, I recommend that anyone who is interested in using speech synthesis services, as I am, pay attention to how a provider has trained their AI when choosing a provider. Reputable providers should be able to provide you with information about their training practices and confirm that they comply with all legal requirements. With this caution, you can ensure that you can take advantage of AI speech synthesis without taking legal risks.
Legal basis for objection
At this point in time, there are several legal bases that individuals can rely on if they want to defend themselves against the use of their voice by AI providers. The focus here is on the right of personality, copyright and the right to privacy.
The right of personality plays a decisive role. It serves to protect individuals from unauthorized use of their image or voice. This right is enshrined in the German Civil Code (BGB) and similar provisions exist in many other countries. Violations of personal rights can have serious consequences, including claims for damages and injunctions. For AI providers, this means that they cannot use individual voices without explicit consent.
Copyright law may also be relevant in this context. Basically, it protects creative expressions from unauthorized uses. An exciting question that arises, however, is whether a voice as such can be considered a work worthy of protection under copyright law. This is a largely unexplored area that could be interpreted differently depending on the country and specific court decision. In any case, AI vendors should be very careful to ensure that they respect the boundaries of copyright law.
Finally, the right to privacy plays a significant role. When creating an AI voice model, a recording of the original voice is usually required. Unauthorized recordings could be considered a serious invasion of privacy, which could possibly result in legal consequences. Therefore, AI providers must ensure that they obtain the necessary consent to record and use a voice.
Overall, it is of utmost importance for AI providers to be mindful of this legal framework and ensure that they fully respect the rights and freedoms of the individuals whose voices they wish to use or model.
The tricky part of this legal question is similar to the question of whether text-generating AI models infringe copyright when they “learn” from the texts of others. Technically, no AI “copies” or rewrites anything directly, but it generates new, independent content based on data it has processed during the training process. This could be described as a form of “hallucinating” new content based on the information the AI has “absorbed” during its training.
It’s a subtle but crucial distinction. AI does not “learn” in the human sense of transcribing or rewriting information, but rather it creates models from the data it receives and uses those models to generate new content. This content is therefore not copies of the original data, but unique products of the AI itself.
However, this feature of AI raises complex legal issues, particularly with respect to copyright. For example, if an AI has been trained with copyrighted text or voices, to what extent is the content it generates affected by that copyright? Since the AI does not directly copy the original data, but only uses it as a basis for generating new content, it can be argued that the generated content does not fall under the copyright of the original data.
However, this interpretation is anything but certain. Copyright is a complex and multifaceted area, and it is unclear how courts will address this issue in the future. It is entirely conceivable that future court decisions could interpret existing copyright laws in ways that restrict the use of AI models in such contexts.
This legal gray area poses a significant challenge for AI providers. It is therefore imperative that they are proactive, seek good advice and take all possible precautions to ensure that they operate within the boundaries of the law. Similarly, it is important that legislators and courts take these issues seriously and create clear legal guidelines for dealing with AI in such contexts.
Compliance for AI SaaS providers
AI SaaS providers must take several steps to ensure they comply with applicable laws. In a world where legal standards for AI and voice cloning are still evolving, vendors, speakers, and all stakeholders should keep a close eye on laws and policies and stay up to date. Here are some actions they should take:
- Obtaining consent: Before using an individual’s voice, the provider should ensure that explicit consent has been obtained from that individual. This consent should be clear, comprehensive and specific to the intended use.
- Respect personal rights: Providers must ensure that they respect the privacy rights of the people whose voices they use. This means that they cannot use personal characteristics such as voice and appearance that make up a person’s personality without consent.
- Ensure transparency: Providers should be transparent about how they use and model voices. This may include providing information about the purpose for which the data will be used, the methods used to collect and process the data, and any third parties who may have access to the data.
In addition, it is important to consider the following aspects:
- Dealing with the legal gray area: AI providers will probably have to operate in a legal gray area for some years to come. You should disclose these circumstances, including to investors and other stakeholders. Anyone claiming otherwise could risk misleading the parties involved.
- Use of general terms and conditions (GTC): GTCs can help protect both users and the providers themselves. Attention should be paid to clauses regulating exemptions and the like. Well-drafted GTCs can also clarify who has what claims against contractual partners in the event that a violation of personal rights or copyright occurs.
- Documentation of the technical basis: The precise documentation of how the AI works and how the voice is technically generated is essential. Should legal proceedings arise in which a software expert opinion is required, this could be a costly and time-consuming process without the appropriate documentation.
It is imperative for AI providers to be proactive, well advised, and take all necessary precautions to ensure they are operating within the confines of the law. This can help minimize potential legal challenges and maximize opportunities for successful operations.