Sonus

In musical practice, woodshedding refers to the solitary, obsessive act of working through technical flaws in private until precision becomes instinct. Sonus is an intelligent polyphonic practice companion for guitarists and pianists. By marrying sub-30ms in-browser chromagram tracking with deep neural multipitch transcription, Sonus acts as an objective auditory mirror — detecting subtle transition lags, dropped chord voicings, and rhythmic rushing before bad habits calcify into muscle memory.

Musicians cannot tolerate audio latency. The core engineering breakthrough in Sonus is its hybrid two-tier pipeline: a dedicated client-side AudioWorklet processor running in a separate audio thread to capture 2048-sample PCM buffers and project energy across 12 pitch classes for instant HUD chord confirmation, combined with asynchronous server-side neural multipitch transcription powered by Spotify's Basic Pitch running on ONNX Runtime. Strum clustering groups rolling guitar strums (35-60ms window) separately from simultaneous piano strikes.

• Sub-30ms real-time audio pipeline using Web Audio API AudioWorklet, avoiding main thread garbage collection pauses.
• Spotify Basic Pitch model executed via ONNX Runtime to produce discrete [pitch, onset, offset, amplitude] note events.
• Needleman-Wunsch global sequence alignment maps user performance against target score exercises, absorbing extra strums or accidental skips without cascading misalignment.
• Adaptive Habit Engine detects transition lag (e.g. F to G > 100ms), tempo rushing (< -30ms on downbeats), and voicing omissions, automatically generating targeted 75% BPM remediation drills.
• Grounded Fact-Bundle Reports compiled strictly from verified mathematical fact bundles (fact_id, category, claim), eliminating speculative advice.

A musician's ear detects auditory feedback lag instantly. Separating immediate feedback (<30ms chromagram) from deep analytical truth (server-side neural multipitch transcription) allowed us to preserve fluid playing flow while retaining analytical rigor.