In November 2022, the company OpenAI (now part of Microsoft) published an online chat system (chatGPT) that generates texts based on instructions (so-called prompts). There are now a wealth of such systems for which the not entirely accurate term “artificial intelligence” (AI) has been popularized. With regard to language and music, AI systems have the potential to automate routine processes in the production and editing of audio material, thus making them very cost-efficient. This development has a radical impact on the fields of activity of sound engineers and other sound creators (see 1 ). The Berlin district group therefore decided to organize a panel discussion to introduce those affected by the topic of AI.

The function of the trained model can be analytically described by a complex mathematical function, the Machine Learning Function (MLF). This MLF can then be used to take decisions or trigger actions, such as classification of music genres, errors in monitored systems, or quality levels of products.



However, generative AI systems require additional processing steps (see e.g. 17 ). The generation of the output of such a system begins with an input text or speech-command that verbally describes the request to the system, the so-called prompt. Prompts can be very detailed. Writing them is a skill in itself. Prompts can even give rise to copyright. The exact processing of prompts is system-specific. In general, the following applies: The system processes the prompt step-by-step. For each step, the system generates an output section based on its learned data. The chain of all these sections results in the overall output.


Systems that do not perform the mixing themselves but instead make suggestions for the parameters of the mixing process, are also attractive. The sound engineers then carry out the actual mixing themselves. This expands their creative scope. It is also possible, for example, to reduce background noise, such as, breathing, reverberation, traffic noise, ventilation noise, crackling.

AI systems for source separation have recently been developed to a commercial level. Separated sources are a prerequisite for object-oriented mixing, e.g., when creating immersive audio and/or constructing virtual auditory spaces.



Finally, the issue was raised by Mr Decker9 and Georg Fett10 as to whether AI-assisted mixing and mastering systems will have an impact on the training of sound professionals. Actually, a trained ear may turn out to be less essential than the ability to operate the AI systems efficiently. Question: Why at all should I then study to become a tonmeister?

Such systems (e.g., Suno, Udio) are very powerful. They set lyrics (your own ones or those generated by ChatGPT) to music and create polyphonic soundtracks in radio quality (192 kBit mp3).


Our colleague Forster Thomas16 commented succinctly: “Now I'm unemployed!”

The training material is usually not disclosed. It is suspected that commercial repertoires of the major music publishers have been tapped for this purpose.

In the prompt, the user gives a few instructions on the genre and emotional expression of the desired composition. The text is then performed by a male or female singer, a male or female choir, or a musical instrument of your choice, and accompanied by a band if desired. With special cloning software (e.g. 19 ) you can use any singing voice – even your own one.



By the way, it may happen that the AI creates a composition that is so close to a natural original that it can be classified as plagiarism. The burden of proof lies with the victims. They can complain to Suno. The system then does not output the AI version, but rather a link to the original on YouTube. This occurs, e.g., with songs by well-known pop stars.

The topic of generative composition systems sparked a lively debate. Peter Hirscher4 started with the frequently asked question of whether AI will ultimately replace humans and suggests considering the following: AI creates many use cases that people would never have thought of. For example, it allows for the creation of purpose-built songs (Scalable Audio) that humans cannot create quickly and inexpensively. This opens the opportunity for creating songs that no one else would produce, but which serve an entertaining purpose.

A thesis which is often put forward is the following:

"Only people are capable of creativity, machines are not."

From this statement, one could deduce that man-made products are more valuable than machine-made ones, which would be a market advantage for their producers. However, is individual art really a market-powerful factor for sound production?

 Simon Hestermann17 writes: “The economic advantage of good AI (KI) cannot be denied. Why should a label 
pay artists to produce music for workout playlists on Spotify when AI delivers 
the right music en masse? Why should a music audience look for artists for a candlelight dinner when AI composes personalized music for the evening? 
Thus, the question is no longer whether, but when technology will present the music industry with a fait accompli. We should not rely on the morality and appreciation of the masses, as the streaming age has already taught us. 
Ultimately, consumer convenience wins!”

This quote was followed by the question of whether only humans are capable of creativity, or whether machines can also be creative? The following thoughts were put forward by Jens Blauert1: AI is known to be able to hallucinate, i.e. imagine something that does not stand up to current fact checking. AI can therefore imagine. Cannot such imagined ideas contain useful suggestions for the future? Isn’t it true that humans do the same thing when getting creative, namely, they start using their imagination?


However, the change triggered by AI has far-reaching social consequences. In short: the massive use of capital by big investors leads to the “expropriation” of individual sound creators7. It is likely that a new social contract is required to tackle this problem in the future4.

The final contributions to this set of questions were the following: Theodor Przybilla11 does not rule this out the advantage of an updated social contract but takes a more optimistic view of the problem. He trusts in the human ability to use new technologies for our benefit and to create adequate conditions.
 Pilos Kostas12 reports that the reception of AI audio products depends on the audience's level of experience. For example, children are often better able to identify AI songs as such than adults (see 20). Might it be that in our society we are permanently and involuntarily drilled to get accustomed to consumer art?

Audio production is already in a transformation phase in which the role of AI is rapidly increasing. As already emphasized in the introductory statement, AI is undoubtedly a useful tool for audio production. However, only those who have mastered its use will be able to use this tool productively. The participants of this panel therefore parted in the certainty that it was necessary to discuss this topic in detail, and this will remain so. The closing statement thus was:

Good that we talked about it!