EchoNotch

Inspiration

Coding or studying with high-pitched ear ringing (tinnitus) and room HVAC rumble is exhausting. Free white-noise websites only play static MP3 loops that keep stimulating your ringing frequency, while clinical notched-noise apps cost over $100/year. Furthermore, a standard text chatbot cannot process live room audio or run real-time digital signal processing (DSP) filters. I built EchoNotch to turn the browser into an AI acoustic engineer and live 60fps sound shield.

What it does

AI Symptom Intake & Hybrid Gemini API Engine: Users select a 1-click clinical preset or describe their ear ringing and room noise in plain English. Users can optionally paste a live Gemini API Key (gemini-2.5-flash) for structured JSON parameter extraction, or leave it blank to run on EchoNotch's built-in deterministic Acoustic NLP & Physics Solver—ensuring 100% zero-friction testing for judges with no API wall.
Frequency Calibration & Mic Scanner: Includes a 200 Hz – 12,000 Hz target slider, a hearing-safe 1.5s reference tone audition, and a live microphone FFT peak scanner (with a simulated telemetry fallback if no mic is connected).
Live Web Audio DSP & 60fps Oscilloscope: Synthesizes procedural Brown ($1/f^2$), Pink ($1/f$), and Velvet noise in RAM (zero MP3 files) routed through a high-pass HVAC filter and dual cascaded BiquadFilterNode notch filters ($Q = \sqrt{2^B}/(2^B - 1)$) to carve a $>36\text{ dB}$ spectral canyon around the user's ringing frequency ($f_0$), paired with optional stereo binaural beats (10 Hz Alpha / 14 Hz SMR).
Offline Therapy Export: Exports a standalone .html acoustic prescription card for offline use.

How Planning (`scope.md`, `prd.md`, `spec.md`) Shaped the Build

Scoping (scope.md): I cut user accounts, cloud databases, and multi-month audiogram charts so I could deliver a complete, single-screen proof of concept in one sitting.
Product & Design (prd.md): Instead of generic light-mode cards, I designed a dark-mode studio oscilloscope UI (#070b12 slate, #00f5d4 cyan trace, #ff3366 crimson notch). Planning edge cases upfront led directly to three key features: an 800ms safe-fade volume ramp to protect ears, a no-mic simulation fallback, and the optional Gemini API key drawer backed by a deterministic local solver.
Specification (spec.md): Mapping out the exact Web Audio API node chain and biquad $Q$-factor math before coding prevented audio routing bugs during implementation.

What I learned

Separating what matters (scope.md), how it feels (prd.md), and how the audio math connects (spec.md) before writing code prevented feature creep and let me build a real-time DSP application in a single afternoon.