Chroma Chips

Twelve-bin chroma features for stabilizing pitch estimates and note labeling. This library demonstrates mixing spectral features with a time-domain pitch guess to reduce octave errors on guitar/voice.

Features

• Compute 12-bin chroma features from FFT magnitude data
• Refine time-domain pitch estimates using spectral information
• Handle frequency ranges from A1 (55 Hz) to C7 (2093 Hz)
• Browser-compatible with TypeScript support

Installation

npm install @playground-sessions/chroma-chips

Usage

import { ChromaFeature } from '@playground-sessions/chroma-chips';

// Initialize with audio context settings
const chromaFeature = new ChromaFeature({
    sampleRate: 44100,
    fftSize: 2048
});

// Compute chroma features from FFT magnitudes
const magnitudes = new Float32Array(1024); // Your FFT magnitude data
const chroma = chromaFeature.computeChroma(magnitudes, Date.now());

// Refine a pitch estimate using chroma features
const pitchGuess = {
    frequency: 440, // Hz
    confidence: 0.5 // 0-1
};

const refinedPitch = chromaFeature.refinePitch(chroma, pitchGuess);
console.log(`Refined pitch: ${refinedPitch.frequency} Hz (confidence: ${refinedPitch.confidence})`);

Demo

Open demo.html in a web browser to try out the library with live microphone input. The demo shows:

• Real-time pitch detection with chroma refinement
• Visual representation of the chroma bins
• Frequency spectrum visualization
• Note name and confidence display

How It Works

1. Chroma Feature Extraction

   • FFT magnitude spectrum is mapped to 12 pitch classes (C through B)
   • Each frequency bin contributes to its corresponding pitch class
   • Features are normalized to sum to 1.0

2. Pitch Refinement

   • Initial pitch guess is converted to a pitch class
   • Chroma features are used to validate or adjust the guess
   • Nearby pitch classes are checked for stronger evidence
   • Confidence is adjusted based on spectral evidence

This approach helps reduce common pitch detection errors, particularly octave errors in complex signals like guitar or voice.

API Reference

ChromaFeature

Constructor

constructor(options: ChromaOptions)

Options:

• sampleRate: Audio sample rate in Hz
• fftSize: Size of the FFT (must match your magnitude data)
• minFrequency: Optional minimum frequency (default: 55 Hz)
• maxFrequency: Optional maximum frequency (default: 2093 Hz)

Methods

computeChroma(magnitudes: Float32Array, timestamp: number): ChromaVector

Computes chroma features from FFT magnitude data.

refinePitch(chroma: ChromaVector, pitchGuess: PitchGuess): PitchGuess

Refines a pitch estimate using chroma features.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT