1. Introduction
In a 1906 essay titled “The Menace of Mechanical Music,” the American composer John Philip Sousa—known for The Stars and Stripes Forever—famously described the sound from the then-proliferating phonograph, one of the earliest recording/reproduction technologies, as “canned music” (Sousa, 1906). Here, Sousa employs the analogy of canned music to contrast the act of listening to recordings with attending live performances, akin to consuming canned fish instead of freshly caught fish. This comparison is set against the backdrop of Sousa’s concerns regarding the potential abandonment of active music-making (performing/catching) by individuals facilitated by the mechanized production of canned music, leading them to assume the role of passive consumers (listening/consuming) (Lessig 2008).
About a century later, Japanese composer Masahiro Miwa, an advocate of “reverse simulation music” (Miwa, 2007), argued that the “music” perceived through the playback of recorded materials should be discussed—or even recognized—as something fundamentally different from music in its traditional sense. To distinguish works experienced through speakers or headphones, he coined the term rokugaku (“recorded music”) and proposed that this be distinguished from the Western musical tradition grounded in the performer’s body and the musical score (Miwa, 2008). While this concept may seem extreme, it departs from the understanding that a “work” receives the “soul” (the performer’s interpretation) anew each time via the medium of music sheets, redirecting attention toward expressions specific to technologies of reproduction.
Jonathan Sterne, a pioneer of sound studies, also draws on early debates on recording and reproduction technologies to argue that “face-to-face” or “live” acoustic events are social practices that are fundamentally different from the acoustic production enabled by technical reproduction (Sterne 2003) and criticizes the distinction between the original and its copy. According to Sterne, the idea of the original as an object of reproduction is impossible without the reproduction process, and recording and reproduction mask the differences between technologies, each with its specific characteristics. Following this line of thought, one might say that the value of freshly caught fish is obscured without canned fish, and the value of live performance is only established in relation to rokugaku.
Despite differences in historical context and aims, these three positions share a communication model in which music (sound) is transmitted from a sender (performer or technology) to a receiver (listener). Framed through Shannon’s notions of “signal/noise” (Shannon 1948), their arguments can be summarized as follows.
For Sousa, just as canning diminishes the freshness of fish, the difference between live and recorded performance represents a form of “noise.” Miwa’s proposal to see live performance and recording as distinct forms of expression might seem to escape the signal/noise opposition at first glance. Yet ontologically divorcing the two raises the question of how to position the various noises involved in the recording/reproduction technology and process in relation to the recording as a signal. Sterne’s assertion that the concept of the original cannot exist without the process of reproduction can be read to suggest that the authenticity of the signal (or, originality) is clarified only through the noises that accompany reproduction.
In these frameworks, what happens if we add sounds synthesized by AI, which has made remarkable progress in recent years? Beginning with WaveNet (Oord et al., 2016), AI techniques for synthesizing audio waveforms through deep learning developed into systems such as Jukebox (Dhariwal et al., 2020), and—by October 2025—prominent examples include Suno (2023) and Udio (2024). These suggest a future in which virtually any sound (music) can be synthesized insofar as it can be represented as digital data inside the computer.
At first hearing, AI-generated sounds that can be appreciated through loudspeakers or headphones without the intervention of the human body seem to be the ultimate form of canned music, or rokugaku, lacking the human soul (interpretation by the performer). One might also take the human-provided prompt text as the “work,” with AI acting as a performer, but this seems merely an attempt to forcibly preserve the Western tradition of the music. The fact that different sounds (music) are “synthesized” each time in accordance with a text (prompt) is qualitatively different from “reproduction” via recording and may be better understood as a purification of only the required signals from among innumerable bits of information. As different sounds and interpretations are produced without bodily performance, they verge on “copies without an original” or “signals without noise.”
The foregoing suggests that AI-based sound synthesis marks the critical point of the sender–receiver transmission model assumed by reproduction theories exemplified by Sousa, Miwa, and Sterne. The technological lineage of sound (music) synthesis and its current culmination call for a different model of communication that cannot be fully described by the Shannon–Weaver model. The next section turns to Gilbert Simondon, whose works have been reappraised in recent media studies and the philosophy of technology, to argue that his perspective, especially developed around 1970 and later collected under the title Communication et Information (2015), offers effective insights into this problem.
Your browser does not meet the minimum requirements to view this website. The browsers listed below are compatible. If you do not have any of these browsers, click on the icon to download the desired browser.
