AI Music: Between Technological Progress and Sonic Limitations

Lately, AI-generated music is appearing more and more frequently on social media platforms and even on Spotify. As someone interested in technology, I find this development quite fascinating. In principle, complete musical pieces are created from white noise - including invented lyrics, different keys and musical styles that can even be mixed together. On YouTube, I’ve discovered really funny interpretations with AI generated music, and I think this technology has its place, especially when you approach it creatively and/or humorously.

However, what I cannot understand is the emotional and sonic approach to this music. Because AI music simply sounds terrible. The sound is comparable to an MP3 file compressed at the lowest quality level. Even when trying to master AI pieces, you can immediately recognize the white noise from which they originated. I find it extremely unpleasant to listen to this music, as it usually sounds terrible due to lack of detail accuracy.

The Reality of Market Flooding

Meanwhile, the entire internet is being flooded with this kind of “music,” making it increasingly difficult to find really good music. Music is supposed to transport emotions, which is exactly what AI is not yet capable of creating today, and possibly never will be.

In 2024, AI actually produced more songs than all human musicians combined - platforms like Mubert alone generated over 100 million tracks in six months. These numbers illustrate the extent of the flooding that is already taking place.

Technical Quality Problems

My perception of poor sound quality is confirmed by other experts. Music producer Thomas Foster describes it aptly: “It’s the technical quality, how the audio material sounds - it’s not at one hundred percent yet, it sounds like a bad MP3.” Professor Dr. Heike Adel-Vu from Stuttgart Media University adds: “But if you listen to the pieces in detail, you find artifacts and there are some things that AI cannot do well.”

The specific problems are diverse: words are hallucinated, syllables are strangely pronounced, parts of a verse suddenly become the chorus. Users report “metallic background noises that sound like a cheap digital reverb plugin from the 90s was used.” When you break down an AI song into individual tracks, often “creepy sound structures” emerge.

Fundamental Limitations

The basic problem is that AI has no understanding of music theory or melody nor emotional transportation. It cannot remain consistent over longer periods and produces voices that sound “artificial like sung into a plastic bucket.” These structural problems cannot be fixed through subsequent mastering, as the artifacts already arise during generation.

The Loss of Artistic Originality

While AI-generated music may initially appear professionally produced, it lacks any originality and specificity. The artistic aspect of AI music has completely disappeared. Instead, the music sounds very stereotypical and boring. AI can replicate existing patterns but fails to create anything genuinely new or emotionally meaningful.

An Ambivalent Conclusion

While AI music can certainly be useful as a tool for inspiration and brainstorming, sound quality, detail accuracy and emotional transport remain significant problems. The technology may be impressive, but the listening experience suffers from these limitations. It becomes interesting when AI music stands unmarked alongside human-produced music - as shown by the case of the band “The Velvet Sundown,” which collected over a million streams before it was revealed that it was completely AI-generated. Laypeople often don’t notice this at all, but every trained listener spotted it immediately.

The flooding of the music market with inferior AI music is no longer a future vision but already reality. While the technology certainly has its justification as a creative tool, we should be aware that it cannot yet achieve the emotional depth and sonic quality of details of human-created music.