Maker compresses song 1,000x to store audio on a single sheet of paper

Maker Makestreme has demonstrated a method for storing music on physical paper by leveraging advanced AI compression. The project, titled "Paper Tunes," successfully reduced a standard audio file to a footprint small enough to be printed as a series of barcodes.

Using Meta's EnCodec, an open-source neural audio codec released in 2022, Makestreme compressed a 2.9MB MP3 song down to approximately 21KB. This represents a compression ratio of roughly 1,000x. The resulting data for the two-minute song was split across eight QR codes, which were printed four per side on a single double-sided sheet of paper. To retrieve the audio, the latent tokens stored in the QR codes must be processed through a specific 24 kHz neural network decoder to reconstruct the original waveform.

The shift to neural audio

Traditional lossy compression formats, such as MP3, cannot achieve this level of reduction because they are already highly optimized. EnCodec operates differently by converting audio waveforms into discrete latent tokens rather than relying on traditional psychoacoustic modeling. This approach allows for extreme data density, enabling complex media to be stored in incredibly small digital footprints.

This experiment echoes early digital storage attempts, such as the 1985 Cauzin Softstrip, which allowed magazines to distribute software via 2D barcodes. In a modern twist, Makestreme also tested the portability of this compressed data by transferring the 21KB file between two ESP32 boards using REYAX RYLR998 LoRa radio modules, proving the data's viability for low-bandwidth wireless transmission.

The cost of efficiency

While the project showcases the efficiency of neural codecs, it also highlights a burgeoning issue of digital fragility. Unlike a vinyl record or a CD, where the data is physically represented in a way that can be read by a compatible mechanical head, the "Paper Tunes" sheet is useless without the exact neural network architecture used to encode it. If the specific decoder is lost or becomes obsolete, the physical paper—which could potentially last for centuries—becomes an unreadable artifact.

What remains

As neural codecs continue to evolve, the ability to store high-fidelity media in minuscule spaces may open new avenues for archival and low-power communication. However, the project leaves open the question of standardization; until these neural decoders are universal, such storage methods remain experimental curiosities rather than practical alternatives to traditional digital media.