AI-Generated Music Floods Streaming Platforms – How to Spot Fakes

Summary

AI music generation technology has improved dramatically over recent months and is now flooding streaming services like Spotify and Apple Music. According to French provider Deezer, already one-third of daily uploaded songs come from AI – approximately 50,000 per day. Most platforms have not implemented independent control mechanisms and instead rely on user self-reporting. For listeners, it is increasingly difficult to distinguish real from artificially generated songs. However, there are several identifying features: generic lyrics with simple rhymes, so-called "ghost instruments" (unexplainable sound artifacts), and subtle errors in instrumentation. Deezer has developed its own algorithm that can reliably distinguish AI music from human music.

People

• Ruth Fulperer
• Simon Schaffel

Topics

• AI music and streaming services
• Identifying features of artificially generated music
• Missing regulation and transparency
• Algorithms for AI detection
• Impact on artists and the music industry

Detailed Summary

The Extent of the Problem

The NZZ has developed an interactive quiz where users must guess whether songs come from humans or AI. The results show that approximately half of readers get each song right – a sign of how difficult the distinction has become. Technology editor Ruth Fulperer explains that most streaming platforms do not publish figures on the spread of AI music. An exception is Deezer, which reveals that already one-third of all uploaded songs are AI-generated – approximately 50,000 pieces daily. For larger platforms like Spotify, a similar percentage is suspected. Some AI songs have even already reached the charts and are listened to millions of times.

Lack of Control and Accountability

Most streaming platforms do require users to indicate whether a song is AI-generated, but they do not conduct independent verification. Spotify focuses primarily on preventing artists from impersonating other artists – actual fraud cases, in other words. According to the company, it is working on an "industry standard" to make the origin more transparent, but has so far not introduced a clear label. YouTube also shifts responsibility to users and does not control this itself. An exception is Deezer, which proudly presents its own solution.

Identifying Features of AI Music

Generic lyrics and simple rhymes: The first and most accessible identifying feature is the song lyrics. AI-generated music typically features extremely generic lyrics with simple end-rhyme patterns. The country song "Walk the Walk" is an example – the rhymes ("Talk," "Walk," "Long") are simple and arbitrary. However, human lyricists can compensate for these weaknesses by editing and improving the AI output. A hobbyist who practices AI music as an art form reported that he spent one to three days refining individual lyrics to create more complex narratives.

Ghost instruments: A more subtle feature is so-called "ghost instruments" – sound artifacts that appear and disappear in AI music. An example: In the song "Walk the Walk," a metallic cymbal-like sound suddenly appears and never occurs again in exactly the same form. This is not an intentional musical element, but an artifact of the generation process. Similarly, unexplained second voices or changing vocal quality throughout the song can occur – comparable to known errors in AI-generated images (such as the notorious six fingers on a hand).

Deezer's Algorithm Solution

Deezer has developed its own algorithm that reliably detects AI music. The key lies in data reduction: music contains approximately 44,000 data points per second – significantly denser than text or images. AI systems must first compress this information to process it. This leads to fine overlays and duplications in the data structure that are inaudible to the human ear but can be recognized by computers. The comparison with a raster image transmission process illustrates this: if squares do not fit together perfectly, denser information arises at the edges – similar to "thicker paint" when manually copying. These interfaces are measurable in AI music.

Interestingly, AI providers could use different algorithms to evade detection, but these produce worse results for the human ear. Therefore, currently all AI music providers and platforms use the same recognizable algorithm because it delivers the best audio quality – an unintended standardization for quality reasons.

Economic Dynamics and Artist Interests

At first glance, one might assume that streaming platforms are interested in AI music because they need to pay artists less. However, major music labels like Universal and Sony have considerable power in the industry. They resist uncontrolled AI generation and threaten to remove their most popular artists from catalogs, leading to negotiations and regulatory pressure.

A major problem is that much AI music is not conceived for real listeners but to be streamed by bots – artists earn money through artificial plays. This is a fraud system that benefits nobody: neither real artists nor the platforms.

Artist reactions are divided. Some see AI music as an additional nuisance in an already difficult business; for them, a clear signal from platforms that they are "taken seriously" would be meaningful. Others are more pragmatic and argue that AI music is a lesser problem than the effects of streaming and the internet in general, or insufficient fair payment by platforms.

Key Takeaways

• AI music generation technology has improved substantially in recent months and is now flooding Spotify, Apple Music, and other platforms with approximately 50,000 new songs daily.
• Most streaming services conduct no independent control and rely on self-reporting – only Deezer offers reliable detection technology.
• AI-generated music can be identified through generic lyrics, simple rhymes, and "ghost instruments" (unexplainable sound artifacts).
• Deezer's algorithm detects AI music by analyzing overlays in the data structure that arise when compressing 44,000 data points per second.
• The current AI music standard is used by all providers because it produces the best audio quality – an unintended standardization that simultaneously enables detection.
• Major music labels like Universal and Sony pressure platforms to limit uncontrolled AI generation.
• Many AI songs are not created for real listeners but to earn money through bot streams – a fraud system that delegitimizes the entire industry.

Metadata