How difficult is real-time single-note detection for piano and woodwinds?

FFT is useful for visualizing the spectrum and can help with note matching, but it is not always the best standalone method for estimating the fundamental frequency. I would prototype with a tuner-style pitch detector first, then use spectral analysis if you need additional information. Also decide early whether the product only checks the played note or whether it must judge timing and note duration too, since timing validation adds another layer of state tracking.

Be careful with microphone processing on phones. Echo cancellation, noise suppression, and automatic gain control can distort the harmonics that pitch detection depends on, so try disabling them when requesting microphone input. You should also smooth results across several frames and apply a confidence threshold; otherwise octave errors and brief background sounds will cause the displayed note to jump around.

For a browser or mobile implementation, the Web Audio API and an AudioWorklet are good starting points. A frame of around 2048 samples at 44.1 kHz gives roughly 45 ms of analysis latency, which is generally acceptable for an educational tool. A YIN-based JavaScript detector such as Pitchy is a reasonable first library. A neural model like CREPE can handle noisier recordings better, but it will use more processing power.

This is very achievable for a first version. Start with monophonic pitch detection using YIN or autocorrelation, then convert the detected frequency into the nearest musical note and compare it with the expected note. The more difficult parts are usually latency, noisy microphones, and deciding how much pitch variation to tolerate. Detecting multiple simultaneous notes is substantially harder, but you do not need to solve that initially.

Piano will probably be more challenging than flute or trumpet. Woodwind and brass notes often have a clearer harmonic structure, while piano strings produce inharmonic partials and decay quickly. Sometimes the fundamental is weaker than the second harmonic, which can lead to octave mistakes. Begin with one instrument and quiet test conditions, then expand your tolerance rules as you collect real recordings.