shiihaa-app/shiihaa-breath-detection

Shiihaa is a breath detection system built into a mobile app, using only the phone's built-in microphone to track your breathing in real time and reflect it back to you as biofeedback. The idea behind it is to create a mindfulness tool that hands attention back to the user rather than competing for it. There is no wearable required, no coach, and no gamification. The core problem it is solving is that a phone microphone in a real room picks up a lot of noise: ambient hum, traffic, a fan, fabric against the phone, automatic gain control from the phone's own software. The system has to separate breathing signal from all of that and identify which phase you are in at any given moment, whether that is inhaling, exhaling, holding, or transitioning between them. It works in three layers. First, the audio stream is cut into short overlapping windows, and each window is analyzed for its energy level and where that energy sits in the frequency spectrum. Inhales tend to be more turbulent and higher in frequency, exhales tend to be lower and smoother. Second, a state machine tracks the current breathing phase and decides when a genuine transition has occurred versus a momentary fluctuation. It uses adaptive thresholds that recalibrate as room conditions change. Third, a quality-check layer discards ambiguous windows rather than making a guess, so the output is honest about uncertainty. Machine learning plays a supporting role, used to sharpen the model over time from quality-checked examples, but the live detection runs on the rule-based pipeline. The README is candid that microphone-only breath detection in uncontrolled conditions is genuinely difficult, and the team is running a validation study against clinical ground truth to measure how well it actually works. All audio processing happens on the device. No raw audio leaves the phone, and the pipeline is designed around breathing patterns specifically, not speech recognition. A browser-based demo lets anyone try the live biofeedback without creating an account. The full app requires an account for personal features such as saved sessions. This is described as a wellness and self-awareness tool, not a medical device, and it does not make diagnostic claims.