Sonic Abyss: Live Sonification of the Hadal Zone

Imagine standing in a gallery in Tokyo, hearing the groan of tectonic plates and the click of a bioluminescent jellyfish from 11 kilometers beneath the Pacific. Sonic Abyss is a project to deploy three self-powered hydrophone nodes at Challenger Deep, the deepest point in the ocean. Each node captures infrasonic whale calls, seismic rumblings, and the subtle pops of deep-sea fauna. The data is processed on a surface buoy by a machine learning model that identifies patterns and maps them to musical parameters—pitch, timbre, rhythm—in real time. The resulting composition is streamed via satellite to a website and a permanent installation, creating a 24/7 generative piece that reflects the heartbeat of the abyss. The project requires custom titanium pressure housings, low-noise hydrophones, acoustic modems, and a robust AI pipeline. It’s a marriage of oceanography, engineering, and algorithmic composition, designed to give the public a visceral connection to a world humans cannot directly experience.

Existing deep-sea exploration is overwhelmingly scientific and inaccessible to the public. Audio recordings from the hadal zone exist but are rarely released as artistic experiences. There is no way for someone to tune into the soundscape of the deepest ocean and hear it evolve over days, weeks, or years. Sonic Abyss aims to fill that gap by transforming raw deep-sea acoustics into a living musical work that is both scientifically informative and artistically compelling. It scratches the itch for a permanent, real-time audio artwork that exists in an extreme environment, making the invisible audible.

• Deploy 3 hydrophone nodes at Challenger Deep (11,000m depth)
• Develop an AI that composes generative music based on real-time environmental data (frequency, amplitude, inter-event intervals)
• Stream audio live 24/7 to a public website and a quiet gallery space
• Ensure the composition never repeats identically, evolving with oceanic activity
• Collaborate with marine biologists to share acoustic data for non-invasive research
• Achieve a system lifespan of at least 6 months without maintenance

• NOT a scientific research project—no peer-reviewed publications are required
• NOT a one-time performance or recording; it is a permanent installation
• NOT using raw ocean sounds as samples—they are algorithmically transformed into tones, pulses, and harmonies
• NOT a commercial product; all streams and code are open-source
• NOT focused on terrestrial or shallow-water environments—only the hadal zone

• Custom pressure-resistant titanium housings (rated to 1100 atm) with sapphire windows
• Low-noise hydrophones (20 Hz – 1 MHz frequency response, -170 dB re 1V/μPa)
• Acoustic modems for inter-node communication (10 kbps at 500m range)
• Surface buoy with 300W solar panels, Iridium satellite transmitter, and a Raspberry Pi 4
• Modified TensorFlow Lite model for real-time detection of events (whale calls, ship noise, bioluminescent clicks)
• Max/MSP for audio synthesis (mapping acoustic features to MIDI parameters)
• Node.js backend for streaming audio via Icecast
• Web frontend with WebSocket for live visualizations

Three hydrophone nodes are deployed in a triangular array, 1 km apart, resting on the seafloor. Each node contains a hydrophone, 24-bit ADC, preamp, an STM32 microcontroller, and an acoustic modem. Nodes communicate raw digitized audio to a central command buoy on the surface via acoustic links (frequency-hopping, 10-50 kHz). The buoy’s Raspberry Pi runs the AI model that classifies sounds and extracts features (peak frequency, amplitude envelope, inter-event intervals). These features are sent via Iridium satellite (compressed, ~2.4 kbps) to a cloud server running Max/MSP. Max/MSP receives the data as OSC messages and uses a predefined mapping: low-frequency content (1-100 Hz) drives drone tones, mid-frequency (100-500 Hz) triggers organic percussion, and high-frequency clicks (1-50 kHz) generate glitchy effects. The composition is output as a stereo stream, encoded to MP3, and broadcast via Icecast to web clients. The gallery installation includes 12 speakers in a circle, a subwoofer, and projected real-time spectral graphs.

• Extreme pressure may cause catastrophic failure of electronics (despite titanium housing)
• Biofouling on hydrophone membranes could degrade sensitivity after months
• Satellite bandwidth insufficient for full-frequency audio—compression may lose nuance
• AI may produce monotonous or uninteresting music without human tuning
• Permits from Papua New Guinea (nearest sovereign nation) could be denied or delayed
• Typhoons can damage the surface buoy and disrupt satellite link

• How to power nodes for >6 months without battery replacement? (Acoustic power transfer? Nuclear thermoelectric? Fuel cells?)
• Can we achieve sufficient bandwidth to stream 44.1 kHz audio, or must we drop to 22 kHz?
• Should the composition be purely generative or allow occasional human intervention (e.g., live mixing by a remote artist)?
• How to prevent the AI from overfitting to repetitive geological hums?
• How to ensure the system can survive a Category 5 typhoon (winds > 250 km/h)?

The project was proposed to a marine research institute but denied funding due to the high cost of custom deep-sea hardware ($2M estimate). The originator moved to Europe for a teaching position and the detailed plans remain on a hard drive. Despite promising prototypes in shallower waters, the momentum was lost—but the design documents are still shared with interested artists.

The AI model was prototyped using NOAA deep-sea recordings from the Marianas. The output was surprisingly musical—melodic drones from tectonic shifts and rhythmic clicks from shrimp. A simplified version could be deployed on off-the-shelf underwater drones (like the SV3) for testing. The project could be revived with newer low-cost titanium 3D printing or carbon fiber housings.