Phone Spirometer App for Noisy Home Use

ISEF Category: Systems Software

Subcategory: Mobile Apps  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

Your phone can hear a whisper across a room, but a fan, TV, or hallway noise can wreck a reading. That makes this project a smart test of signal processing, not just app building. You get to ask a real question, can a phone still estimate breathing strength when the home is messy and loud?

What Is It?

This project turns a phone microphone into a rough breathing sensor. You record a forced exhale, then use software to clean up the audio and estimate peak expiratory flow, which is the highest speed of air coming out during a strong breath. Think of it like trying to measure the splash from a pebble in a pond while raindrops are hitting the water too. The app has to find the breathing signal inside the noise.

The key idea has two parts. Denoising removes unwanted sound, like a filter that keeps the main melody and lowers the background hum. Regression means the app learns a rule that maps features from the audio, like loudness shape or frequency patterns, to the meter reading. You then compare the phone estimate with a low-cost peak-flow meter to see how close the app gets.

Why This Is a Good Topic

This is a strong science fair topic because you can test clear variables, like room noise level, microphone placement, and denoising method. It connects to a real health problem, home breathing checks for asthma or other respiratory conditions, where cheap tools matter. You can learn signal processing, model evaluation, and error analysis without needing a hospital lab.

Research Questions

• How does background noise level affect the accuracy of phone-based peak expiratory flow estimates?
• What is the effect of denoising method on agreement between microphone readings and a mechanical peak-flow meter?
• Does microphone distance from the mouth change the model error for peak expiratory flow prediction?
• To what extent does model type improve prediction compared with a simple loudness-based baseline?
• Which audio features best predict peak expiratory flow in a noisy home setting?
• How does training on one room condition affect performance in a different room condition?

Basic Materials

• Smartphone with microphone access and audio recording app or custom app
• Low-cost mechanical peak-flow meter
• Quiet room and one noisy room for comparison
• Laptop with Python installed
• Notebook for labeling trial conditions
• Tape measure for consistent microphone placement
• Optional external phone stand or tripod.

Advanced Materials

• Reference spirometry or calibrated peak-flow device if available
• External USB microphone for comparison runs
• Acoustic calibration tone or reference recorder
• Laptop with Python, NumPy, pandas, SciPy, scikit-learn, and librosa
• Audio interface for controlled recording setup
• Room noise meter or sound level meter
• Access to a small pilot participant pool with consent.

Software & Tools

• Python: Runs audio processing, feature extraction, and regression models for your dataset.
• librosa: Extracts audio features such as loudness, spectral shape, and timing.
• scikit-learn: Trains and compares regression models and cross-validation pipelines.
• ImageJ: Helps inspect screenshots, plots, or calibration visuals if you document setup images.
• Audacity: Lets you listen to recordings and inspect noise, clipping, and signal quality.

Experiment Steps

1. Define the signal you want to predict and the app output you will score against the peak-flow meter.
2. Choose the noise conditions you will test first, then keep them consistent across all trials.
3. Plan a recording setup that fixes phone position, breathing prompt, and meter use so your comparisons stay fair.
4. Build a baseline model before adding denoising, so you can measure the value of each processing step.
5. Design your validation plan around separate training and test sets, not the same recordings used twice.
6. Decide how you will report error, bias, and agreement, so your result shows more than a simple correlation.

Common Pitfalls

• Letting the phone clip during strong exhalations, which destroys the audio features you need for prediction.
• Changing microphone distance between trials, which makes the model learn position instead of breath strength.
• Training and testing on the same people, which makes the app look better than it will on new users.
• Using only quiet-room recordings, which hides how badly the model fails in real homes.
• Comparing predictions to the wrong meter reading, which breaks the ground truth for the whole project.

What Makes This Competitive

A stronger project goes beyond "does it work" and asks when, why, and for whom it works. You can raise the level by testing multiple noise types, comparing several denoising methods, and using a held-out participant set the model never saw during training. Strong error analysis matters here, especially if you report whether the app overestimates or underestimates breath strength under specific conditions. That kind of careful validation makes the project feel like real engineering.

Project Variations

• Test whether the app works better with children, teens, or adults by comparing voice and breath signal patterns across age groups.
• Compare denoising on-device versus denoising after recording to see which workflow gives better predictions in noisy rooms.
• Swap the peak-flow meter target for a different respiratory metric, such as forced exhale duration or breath sound intensity.

Learn More

• NIH PubMed: Search for review articles on peak expiratory flow, home spirometry, and respiratory sound analysis.
• NCBI Bookshelf: Read free biomedical background chapters on spirometry and pulmonary function testing.
• NOAA Sound Level Basics: Learn how background noise is measured and why it affects recordings.
• MIT OpenCourseWare: Find free signal processing and machine learning lecture materials for feature extraction and regression ideas.
• scikit-learn User Guide: Use the free documentation to compare regression models and evaluation methods.
• Python librosa Documentation: Learn how to extract audio features from microphone recordings.