This guide walks you through real-time pitch detection from two common sources:

1. 🖊 Microphone / Line-in audio via the Web Audio API (JavaScript), PyAudio + aubio (Python), and PortAudio + aubio (C / C++)
2. 🎹 MIDI note messages – receiving events and translating MIDI note numbers to frequencies

For each stack you’ll find:

• Signal acquisition   ➟ mic permission / MIDI access
• Pre-processing       ➟ down-mixing, windowing, gating
• Pitch algorithms – Zero-Cross, Autocorrelation, FFT, YIN
• Libraries & example code (ES5-compatible JS + Node; Python 3; C/C++)
• Parameter tables so you know what every call expects

1.1 Signal Chain

1. getUserMedia({audio:true}) ➝ MediaStream
2. new AudioContext()
3. createMediaStreamSource(stream)
4. ScriptProcessorNode (broadest support) or AudioWorklet
5. Frame → Float32Array ➝ Pitch algorithm ➝ frequency (Hz)

1.2 Core Code (ES5-friendly)

// ES5 – 100 % legacy-safe
var BUF = 2048;           // analysis hop-size
var AMP_GATE = 0.006;     // ≈ -45 dBFS
var audioCtx, procNode, yin;

(function init(){
  // ① import YIN polyfill
  var s = document.createElement('script');
  s.src = 'https://cdn.jsdelivr.net/npm/pitchfinder@2.3.0/dist/pitchfinder.iife.js';
  s.onload = askMic; document.head.appendChild(s);
})();

function askMic(){
  navigator.mediaDevices.getUserMedia({audio:true}).then(function(str){
    audioCtx = new (window.AudioContext||webkitAudioContext)();
    var src  = audioCtx.createMediaStreamSource(str);
    procNode = audioCtx.createScriptProcessor(BUF,1,1);
    yin      = Pitchfinder.YIN();           // defaults: 44100 Hz
    src.connect(procNode);
    procNode.connect(audioCtx.destination); // gain-0 hack optional
    procNode.onaudioprocess = onAudio;
  }).catch(function(e){ alert(e); });
}

var buf = new Float32Array(BUF);
function onAudio(ev){
  var data = ev.inputBuffer.getChannelData(0);
  buf.set(data);
  if(rms(buf) < AMP_GATE) return;
  var hz = yin(buf);
  if(hz) console.log('🎵',hz.toFixed(2),'Hz');
}
function rms(a){var s=0;for(var i=0;i<a.length;i++)s+=a[i]*a[i];return Math.sqrt(s/a.length);}

1.3 Important Parameters

Parameter / Method	Description	Typical Range
BUF	Samples per analysis frame (hop-size)	512 – 4096
AMP_GATE	RMS gate; ignore very quiet audio	0.003 (≈-50 dB) – 0.03 (≈-30 dB)
Pitchfinder.YIN(opts)	opts.threshold (dip), opts.probabilityThreshold (quality)	0.1 – 0.2 / 0.8 – 0.95

1.4 Alternative Algorithms

• Zero-Crossing – Count sign changes; simplest, least accurate.
• Autocorrelation – Compare signal with delayed versions; stable for monophonic.
• FFT peak-picking – Good for tonal sounds; latency ≈ FFT size.
• Machine Learning (CREPE, Onsets-Frames) – heavy, needs WASM / TensorFlow.js.

1.5 Python 3 Example (PyAudio + aubio)

import pyaudio, aubio, numpy as np, time
BUFFER = 2048
p      = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paFloat32, channels=1,
                rate=44100, input=True, frames_per_buffer=BUFFER)
pitch_o = aubio.pitch("yin", BUFFER*2, BUFFER, 44100)
pitch_o.set_unit("Hz"); pitch_o.set_silence(-40)
print("Listening… Ctrl-C to stop"); time.sleep(.3)
try:
  while True:
    data = np.frombuffer(stream.read(BUFFER,exception_on_overflow=False),
                         dtype=np.float32)
    hz = pitch_o(data)[0]
    if hz > 0: print(" %.2f Hz" % hz)
except KeyboardInterrupt:
  stream.stop_stream(); stream.close(); p.terminate()

1.6 C / C++ Skeleton (PortAudio + aubio-C)

// compile with: gcc mic_pitch.c -laubio -lportaudio -lm
#include <aubio/aubio.h>
#include <portaudio.h>
/* init PortAudio, open stream, call aubio_pitch_do in callback */

2.1 MIDI Basics

A note-on message [0x9n, key, vel] carries a MIDI note number (key 0 – 127). Convert to frequency via:

f (Hz) = 440 * 2^((key - 69) / 12)

2.2 Web MIDI API (ES5)

if(navigator.requestMIDIAccess){
  navigator.requestMIDIAccess().then(function(acc){
    var inputs = acc.inputs.values();
    for(var inp of inputs)
      inp.onmidimessage = function(e){
        var d = e.data;              // Uint8Array(3)
        if((d[0] & 0xf0) === 0x90 && d[2]!==0){   // note-on
          var note = d[1];
          var hz   = 440 * Math.pow(2,(note-69)/12);
          console.log('🎹', note, '->', hz.toFixed(2), 'Hz');
        }
      };
  });
}

2.3 Node / Electron ( midi or easymidi )

const midi = require('easymidi');
const input = new midi.Input('IAC Driver Bus 1');
input.on('noteon', msg => {
  const hz = 440 * Math.pow(2,(msg.note-69)/12);
  console.log(`Note ${msg.note} ➞ ${hz.toFixed(2)} Hz`);
});

2.4 Python 3 (mido + python-rtmidi)

import mido, math
with mido.open_input() as port:
  for msg in port:
    if msg.type == 'note_on' and msg.velocity:
      hz = 440 * 2 ** ((msg.note-69)/12)
      print(msg.note, '->', f'{hz:.2f} Hz')

2.5 Parameter Reference

Field	Description	Value range
0x9n	Status byte (note-on, channel n)	0x90 – 0x9F
key	MIDI note number	0 – 127
velocity	Loudness; 0 = note-off shortcut	0 – 127

• Buffer size: 2048 @ 44.1 kHz ≈ 46 ms analysis window – trade-off between latency & low-frequency resolution.
• Low-freq detection: For bass (≤ 60 Hz) use BUF ≥ 4096 or down-sample to 22.05 kHz.
• Dynamically adapt gate: set AMP_THRESHOLD to 4 × ambient RMS measured while idle.
• Debounce pitch output: use an exponential moving average (SMOOTH_ALPHA ≈ 0.25) to tame jitter.
• Browser permissions: mic access requires HTTPS or localhost in Chromium; Web MIDI requires a user gesture in Chrome ≥ 94.
• Unity / game engines: route mic via AudioSource ➞ GetSpectrumData, or embed aubio C library for < 10 ms latency.

Q 1 · Why does YIN sometimes return 0 or NaN?

The frame failed its periodicity probability test. Lower probabilityThreshold or pad with zeros for FFT methods.

Q 2 · How can I detect chords?

Use Constant-Q Transform or Pitch Class Profile (lib: meyda JS, librosa Python) then pattern-match triads.

Q 3 · Why is Web MIDI silent in Safari?

Safari (2025-05) still flags Web MIDI behind Develop → Experimental Features → MIDI – System Exclusive. Enable and reload.