Critical Assessment
The TAM app has been finalised and released on the App Store, with a dedicated website at theaquaticmuseum.com. A complete critical assessment of the app will benefit from real-world feedback now that it is publicly available. Online polls and small focus groups can provide further insights into usability questions. In the meantime, I’ll reflect on the creation process and the exchanges with the musicians, and give some personal reflections on the potential audience of music apps. In the process section, some of the following points have been discussed in more detail.
Creator perspective
Adaptive audio is here to stay and offers challenges and opportunities for the composer and performer, moving away from the linear constraints of composing music. The aleatory, chance-related functions and features will become powerful assistants in composing and performing.
Research led me to explore Karnatic development; although it will take months and years (I’ll add a second year of lessons in 2022/2023) to gain a solid grasp of the theory and possibilities, the groundwork is there to apply this knowledge to generative electronic and more exploratory music.
Other tools would have led to more experimental results; for example, MAX by Cycling ‘74 or Pure Data are designed for advanced audio manipulation, but the FMOD audio engine combined with the Unity Game engine guaranteed my ability to release a finished product. MAX packages and apps can be created and distributed, but they don’t reach the user experience of native apps and, most importantly, they cannot be ported to mobile devices or the internet at this time. The Unreal game engine with sequencing possibilities and evolution on iOS is interesting because it brings a set of synthesis and audio manipulation tools.
To put it very bluntly: “Is all this effort for a 3-minute pop song worth it?” Considering the way pop music is produced1 today on an industrial scale with industrial processes, this question is worth asking.
The potential of generative music is also genre-dependent. Brian Eno’s ethereal music is probably more suited to the generative music process than groove-based music.
It was difficult, and in some cases even impossible, to break out of the loop framework. Advanced rhythmic techniques require re-recording of every instrument except when opting for electronic sounds only. Despite these inherent limitations of the FMOD tool, the TAM app is a unique opportunity to explore all available options for one simple reason: in this project, I am not limited by heavy audio compression or by having to worry about performance issues with sound and music. In standard video games, which includes 99.9% of them, game performance comes first. Audio and sound are ranked second, which means using heavy compression to reduce file size. This is not the case here because the visual media content consists of a three-minute hand-drawn video with minimal color information. The interactive parts are all timed, and the user has only a short window of opportunity to interact with the material.
Copyright is also a domain that hasn’t been tackled in this research. The question of registering the Souvenir Shop song is still an open one.
Pertinence for the TAM Project
Audience Perspective
One open question concerning the audience and the app as a marketing tool is: Will the app motivate the audience to listen repeatedly to the same 3-minute Souvenir Shop tune? Will they notice the differences?
The originality of the TAM app also reveals its weak point. Today’s audiences have high expectations when playing video games or interacting with apps. This is one reason to limit interactions to a few seconds by removing options.
Summary
To conclude, I’ll add the following points:
-
Now that the app has been released, a complete critical assessment from the audience perspective can draw on real-world feedback.
-
Other tools would have led to more experimental results; for example, MAX by Cycling ‘74 or Pure Data. However, publishing format is key.
Despite these limitations, this research was a unique opportunity to explore all available options: in this project, I am not limited by heavy audio compression or having to worry about performance issues with sound and music.
- Research led me to explore Karnatic development and experiment with an electronic music approach; although it will take more time to gain a solid grasp on the theory and possibilities, the groundwork is there to apply this knowledge to generative electronic and more exploratory music. Live coding is next.
Postscript: The Rise of AI-Generated Music
Since this research was completed, the landscape of generative music has shifted dramatically with the emergence of large-scale AI music models. Platforms such as Suno and Udio can generate full arrangements — vocals, lyrics, and instrumentation — from text prompts alone, while Google DeepMind’s MusicLM (Agostinelli et al., 2023) and Meta’s MusicGen (Copet et al., 2023) have demonstrated high-fidelity generation conditioned on text, melody, or both. These tools represent a fundamentally different paradigm from the rule-based and stochastic approaches explored in this thesis: where TAM relies on carefully authored musical material recombined through adaptive audio middleware, neural music generation produces material ex nihilo from learned statistical distributions.
This distinction matters. The compositional decisions embedded in TAM — voicing choices, Karnatic rhythmic structures, metric modulations — reflect deliberate musical craft. AI-generated music, by contrast, raises unresolved questions about authorship, originality, and the role of the composer (Sturm et al., 20192). The copyright concerns noted earlier in this chapter have only intensified: at the time of writing, lawsuits from major record labels against AI music platforms are ongoing, and the legal status of AI-trained-on-copyrighted-material remains unsettled.
Nevertheless, the convergence is worth noting. Future iterations of projects like TAM could potentially combine hand-crafted adaptive structures with AI-generated material, using models not as a replacement for compositional intent but as another source of controlled surprise — yet another king or queen of Serendip.
