Python  

Compute FFT for Audio Pitch Detection: Real-Time Guitar Tuner in Python

Table of Contents

  • Introduction

  • What is FFT and Why It Matters for Pitch Detection

  • Real-World Scenario: Building a Real-Time Guitar Tuner

  • Step-by-Step FFT Implementation for Pitch Detection

  • Complete Working Code with Live Audio Input

  • Performance Tips and Best Practices

  • Conclusion

Introduction

Detecting the pitch of a musical note in real time might sound like magic—but it's pure signal processing powered by the Fast Fourier Transform (FFT). Whether you're building a guitar tuner, a voice assistant, or an AI music analyzer, FFT is the cornerstone algorithm that turns raw audio into actionable frequency data.

In this article, you'll learn how to implement FFT-based pitch detection in Python using only standard scientific libraries—and apply it to a real-time electric guitar tuner, a practical tool musicians use daily.

What is FFT and Why It Matters for Pitch Detection

The Fast Fourier Transform (FFT) is an efficient algorithm to compute the Discrete Fourier Transform (DFT), which converts a time-domain signal (like microphone input) into its frequency-domain representation.

In simple terms:

  • Time domain: Amplitude vs. time (what your microphone records)

  • Frequency domain: Amplitude vs. frequency (what your ear perceives as pitch)

For pitch detection, we're interested in finding the dominant frequency—the strongest peak in the FFT output—which usually corresponds to the fundamental frequency of the played note.

Real-World Scenario: Building a Real-Time Guitar Tuner

Imagine you're on stage, your guitar goes slightly out of tune between songs, and you need instant feedback. Commercial tuners do this silently—but you can build your own live Python tuner that listens through your laptop's microphone and tells you if your E-string is sharp, flat, or perfect.

This isn't hypothetical: musicians in home studios, educators, and app developers use similar logic in real products. Our implementation will:

  • Capture live audio from the microphone

  • Compute FFT on small chunks of audio

  • Detect the strongest frequency

  • Map it to the nearest musical note (e.g., A4 = 440 Hz)

  • Display tuning status in real time

PlantUML Diagram

Step-by-Step FFT Implementation for Pitch Detection

1. Capture Audio

Use pyaudio to stream audio in real time.

2. Preprocess the Signal

  • Apply a window function (e.g., Hann window) to reduce spectral leakage

  • Ensure consistent sample rate (e.g., 44.1 kHz)

3. Compute FFT

Use numpy.fft.rfft for real-valued input (faster and sufficient for audio).

4. Find Dominant Frequency

Locate the index of the maximum magnitude in the FFT result and convert it to Hz.

Complete Working Code with Live Audio Input

PlantUML Diagram
import numpy as np
from scipy.io import wavfile
import math
from google.colab import files
import io

# Constants
SAMPLE_RATE = 44100      # Hz (Standard sample rate for the analysis logic)
NOTE_NAMES = ['C', 'C#', 'D', 'D#', 'E', 'F', 'F#', 'G', 'G#', 'A', 'A#', 'B']

def freq_to_note(freq):
    """Convert frequency in Hz to nearest musical note (e.g., 'A4')"""
    if freq <= 0:
        return "Silence"
    # A4 = 440 Hz = MIDI note 69
    midi_note = 69 + 12 * math.log2(freq / 440.0)
    midi_note = round(midi_note)
    
    # Ensure calculation is robust for extremely high/low notes
    if midi_note < 0 or midi_note > 127: # Limit to standard MIDI range
        return "Out of Range"
        
    note = NOTE_NAMES[midi_note % 12]
    octave = midi_note // 12 - 1
    return f"{note}{octave}"

# ----------------------------------------------------------------------

def detect_pitch(audio_chunk, sample_rate):
    """Detect dominant pitch from audio chunk using FFT"""
    if len(audio_chunk) < 2: # Need at least 2 samples for FFT
        return 0.0

    # Ensure audio data is float for numpy processing
    if audio_chunk.dtype != np.float32:
        audio_chunk = audio_chunk.astype(np.float32) / np.max(np.abs(audio_chunk)) # Normalize and convert

    # Apply Hann window to reduce spectral leakage
    windowed = audio_chunk * np.hanning(len(audio_chunk))
    
    # Compute real FFT
    fft_result = np.fft.rfft(windowed)
    fft_magnitude = np.abs(fft_result)
    
    # Find peak frequency index (excluding the DC component at index 0)
    # Start search from index 1
    peak_index = np.argmax(fft_magnitude[1:]) + 1 
    
    # Convert index to frequency (Hz)
    freq = peak_index * sample_rate / len(audio_chunk)
    
    # Filter out very low/high frequencies (likely noise or harmonics)
    # Standard guitar range: ~82 Hz (E2) to ~1318 Hz (E6)
    if freq < 60 or freq > 1500: 
        # Optionally, try to find the next most dominant frequency or use harmonic analysis
        # For simplicity, we just filter it out here.
        return 0.0
        
    return freq

# ----------------------------------------------------------------------

def main_colab_file_based():
    """Main function to run in Colab, processing an uploaded file."""
    print(" Colab Guitar Tuner (File-Based)")
    print("Please upload a .wav file containing a single, clear guitar note.")
    
    # Upload file dialog
    uploaded = files.upload()
    
    if not uploaded:
        print("\nNo file uploaded. Exiting.")
        return

    # Assuming only one file is uploaded
    filename = next(iter(uploaded))
    print(f"File '{filename}' uploaded.")

    try:
        # Read the WAV file data
        samplerate, data = wavfile.read(io.BytesIO(uploaded[filename]))
        
        if samplerate != SAMPLE_RATE:
             print(f" Warning: File sample rate ({samplerate} Hz) differs from analysis rate ({SAMPLE_RATE} Hz).")
             # Resampling is complex, so we'll use the file's rate for calculation but warn the user.
             analysis_rate = samplerate 
        else:
             analysis_rate = SAMPLE_RATE
             
        # Convert stereo to mono by taking the average, if necessary
        if data.ndim > 1:
            data = data.mean(axis=1)

        # Use only the first segment of the audio for pitch detection
        # This acts like the CHUNK_SIZE/Pyaudio buffer in the original code
        CHUNK_SIZE_FOR_ANALYSIS = int(0.5 * analysis_rate) # Analyze 0.5 seconds of audio
        audio_chunk = data[:CHUNK_SIZE_FOR_ANALYSIS]

        if len(audio_chunk) == 0:
            print("Error: Audio chunk is empty.")
            return

        print(f"\nAnalyzing {len(audio_chunk)/analysis_rate:.2f} seconds of audio...")
        
        # Detect pitch
        freq = detect_pitch(audio_chunk, analysis_rate)
        
        if freq > 0:
            note = freq_to_note(freq)
            print("---------------------------------------")
            print(f" Detected Note: {note}")
            print(f"   Frequency: {freq:.2f} Hz")
            print("---------------------------------------")
        else:
            print("---------------------------------------")
            print(" No clear pitch detected (or frequency outside the guitar range 60-1500 Hz).")
            print("   Try uploading a clearer recording of a single note.")
            print("---------------------------------------")
            
    except Exception as e:
        print(f"\nAn error occurred during file processing: {e}")

if __name__ == "__main__":
    # You might need to install scipy first in a Colab cell: !pip install scipy
    # Run the file-based main function
    main_colab_file_based()
34

Performance Tips and Best Practices

  • Use appropriate chunk size: Too small → noisy FFT; too large → laggy response. 1024–4096 works well for real-time.

  • Apply windowing: Always use a window (Hann, Hamming) to minimize edge artifacts.

  • Filter irrelevant frequencies: Ignore <50 Hz (rumble) and >2 kHz (for guitar) to avoid false peaks.

  • Debounce output: Add a small delay or smoothing to prevent flickering note labels.

  • Consider autocorrelation: For very low notes or polyphonic audio, time-domain methods like autocorrelation may outperform FFT.

Conclusion

You've just built a real-time guitar tuner using FFT—no external APIs, no black-box libraries. This same technique powers voice assistants, music transcription apps, and even scientific instruments. FFT isn't just theory; it's a practical, powerful tool that turns sound into insight. Whether you're tuning a guitar, analyzing bird calls, or detecting engine faults, mastering FFT unlocks a world of audio intelligence. Now plug in your guitar, run the code, and never be out of tune again!

Pro Tip: Extend this to detect tuning deviation (e.g., “A4: -5 cents”) by comparing detected frequency to the ideal note frequency—perfect for precision tuning!

Membership not found