From keystroke to soundwave — the technology that makes music respond to you in real time.
TeraMuse's adaptive engine operates on three layers: input analysis, compositional logic, and audio rendering. The input layer captures your keystroke timing, rhythm patterns, and activity state. The compositional layer maps these inputs to musical parameters — tempo, arrangement density, harmonic tension, and timbral brightness. The rendering layer assembles the final audio in real time from pre-composed musical elements stored in the .MUSE file. The entire pipeline operates with latency under 50 milliseconds, ensuring the music feels instantaneously responsive to your actions.
TeraMuse monitors your keystroke timing at the operating system level, extracting several musical parameters: average keystrokes per minute (overall pace), inter-keystroke intervals (rhythm pattern), typing burst length (work period duration), and pause duration (break detection). These raw inputs are smoothed using rolling averages to prevent the music from jerking with every single keystroke. The smoothing window is typically 3–5 seconds, long enough to filter noise but short enough to track genuine pace changes. The system also detects higher-level patterns like acceleration, deceleration, and rhythm regularity.
A .MUSE file is not an audio recording — it's a structured composition containing multiple arrangement layers, transition rules, and parameter mappings. Each file contains stems (individual instrumental parts) at multiple intensity levels, crossfade points where the engine can seamlessly switch between arrangements, tempo elasticity ranges defining how far the track can stretch without artifacts, and behavioral rules governing how the music responds to input changes. A typical .MUSE file contains 50–200 individual audio stems and dozens of transition points, giving the engine enormous combinatorial freedom while ensuring every possible arrangement sounds musically coherent.
The rendering engine continuously assembles audio from the .MUSE file's stems based on current input parameters. It uses horizontal re-sequencing (choosing which section to play next) and vertical layering (choosing which stems to activate simultaneously) to create the moment-to-moment arrangement. Tempo adjustment uses high-quality time-stretching algorithms that preserve pitch and timbre across a ±30 BPM range. Crossfades between arrangement states use psychoacoustically optimized curves that mask the transition within the existing musical texture. The result is a continuous, seamless audio stream that sounds like a single intentional performance.
No. TeraMuse captures only keystroke timing data — the intervals between key presses — not the actual keys pressed. It has no knowledge of what you're typing, only how fast and how rhythmically. No text content is ever captured, stored, or transmitted. The timing data is processed locally in real time and immediately discarded after use. This is architecturally identical to how a typing speed test works — it measures cadence, not content.
Most users notice the adaptation within minutes. The clearest moments are transitions: when you stop typing and the music softens, or when you resume and it builds back up. During sustained typing, the adaptation manifests as the music feeling 'right' — matching your pace in a way that static recordings never do. After a few sessions, switching back to a regular playlist feels noticeably disconnected, because you've become accustomed to music that moves with you.