AI pairing voices and the problem of identifying violations
The development of artificial intelligence (AI) allows the creation of music products from aggregated data. In which, voice is not only imitated and modeled according to a specific recording but also synthesized from many different sources.
From the perspective of an insider, musician Ha Anh said: "Listeners can recognize the familiar voice color, but cannot accurately identify who it is". According to him, when voices are synthesized from many sources, artists can feel their personal imprints are affected, but there is not enough basis to confirm that the product uses their voice.
From a legal perspective, proving violations in these cases faces many obstacles. Lawyer Pham Quoc Bao analyzed: "Firstly, the voice is only indirectly protected through the rights of the performer (Article 29 of the IP Law) and the rights of the recording producer (Article 30) - meaning it must be proven that AI has used the specific original recording of the artist. If AI only "learns" from many small sources (fragmented data), it is difficult to prove that this is a direct copy or derivative work.
Second, AI is often trained on millions of voice samples from many different artists, leading to "voice grafting" to create new products, not exactly like any recording. The identification of violations cannot only be based on similarity, but must be based on specific technical basis. However, with the operating method of AI, this factor is difficult to prove in practice.
Barriers in practice
The amended Law on Intellectual Property in 2025 is promulgated to create a clearer legal basis in the context of rapid technological development. However, according to experts, practical application still faces many difficulties.
According to Lawyer Pham Quoc Bao, although the regulations in the amended law are clear, it is still not easy to apply in practice. One of the obstacles is determining the level of human participation in products with AI elements.
The lawyer analyzed: "In the digital age, the boundary between authors using AI as a supporting tool and allowing AI to completely replace creative thinking is extremely fragile. So how to determine the rate of human intervention in a hybrid work? This is a difficult question for competent authorities to determine whether the conditions for intellectual property registration are sufficient.
In addition, the implementation of the law also depends on guiding documents. "The law is just a framework, many important contents (critics for protecting AI-assisted works, conditions for using AI training data, pricing mechanisms and IP transactions) must wait for the Government's Decree. To date, the system of identification, digital work depository and collective rights management organization is still weak, making the transparency of origin and royalty collection very difficult to implement" - the lawyer said.
In practice, copyright infringement still occurs. Lawyer Pham Quoc Bao commented: "Although there is a clear legal framework, copyright infringement in music, theater, and cinema still occurs a lot.
Lawyers pointed out reasons such as taking advantage of loopholes in electronic websites, content easily copied and spread but also quickly erased traces; users' habit of using free content... Meanwhile, the lengthy litigation process makes many authors discouraged.
The burden of compliance costs is also a barrier. To protect works in the digital environment, authors must carry out many procedures from custody, copyright registration to monitoring and handling violations, while violations occur quickly and widely.
The collection and preservation of evidence also face difficulties when the violating content may be changed or removed in a short time. This prolongs the processing process and reduces the effectiveness of protecting the rights of the author.
