Identify_Chords

Identifying Musical Chords Using FFT

This project identifies the musical notes present in audio recordings of chords using frequency-domain analysis. The core idea is to apply the Fast Fourier Transform (FFT) to .wav files, detect dominant frequencies, and map those frequencies to the nearest musical notes.

Overview

A musical chord consists of multiple notes played simultaneously. Each note corresponds to a fundamental frequency. By transforming the time-domain audio signal into the frequency domain, we can identify these frequencies and infer the notes that form the chord.

This repository handles two cases:

• Long-duration chords (clear spectral peaks)
• Short-duration chords (challenging due to poor frequency resolution)

Methodology

1. Read Audio Data

Audio files are read using scipy.io.wavfile, yielding the sampling rate and waveform.

from scipy.io import wavfile

fs, signal = wavfile.read("short3.wav")

2. FFT Analysis

For long signals, a standard FFT is sufficient. For short-duration chords, the signal is zero-padded to improve frequency resolution:

N_fft = 4000
signal_padded = np.zeros(N_fft)
signal_padded[:len(signal)] = signal

fft_values = np.fft.fft(signal_padded)
fft_magnitude = np.abs(fft_values[:N_fft // 2])

Zero-padding does not add information, but it interpolates the FFT, making peak detection more reliable.

3. Peak Detection

Dominant frequencies are extracted using peak detection:

from scipy.signal import find_peaks

peaks, _ = find_peaks(
    fft_magnitude,
    height=0.1 * np.max(fft_magnitude),
    distance=5
)
frequencies = peaks * (fs / N_fft)

4. Note Identification

Detected frequencies are mapped to the nearest known musical note using a predefined frequency table (octaves 2–5):

idx = np.argmin(np.abs(all_freqs - detected_frequency))
note = all_notes[idx]

Duplicate notes are removed to obtain the final chord.

Example Output

Below is an example FFT magnitude spectrum for a short chord, with detected notes highlighted:

Limitations & Improvements

Short-duration signals suffer from poor frequency resolution due to the time–frequency tradeoff.

Possible improvements:

Apply a window function (Hann or Hamming) before FFT

Use STFT or spectrogram-based averaging

Increase robustness using harmonic grouping

Use parabolic interpolation around FFT peaks