Microtonal Pitch Correction for Maqam Vocals

ISEF Category: Technology Enhances the Arts

Subcategory: Music and Image Manipulation  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Most pitch correction tools aim for 12-note Western tuning. Maqam music often needs notes that sit between those piano keys. That means a normal autotune-style plugin can flatten the style instead of supporting it. Your project asks a sharper question, can software help a singer stay in tune without erasing the microtonal feel?

What Is It?

This project builds a real-time audio plugin that listens to a singer and nudges pitch toward maqam quartertones. Maqam is a system of melody modes used in Arabic music, and some of its notes fall between the notes on a piano. Think of it like a map with extra stops that most Western instruments do not mark. Your plugin tries to guide the voice toward those stops without squashing the small pitch wiggles that make a voice sound alive.

The technical core has two parts. First, pitch tracking estimates the singer’s current note. CREPE is a neural network model that predicts pitch from audio, so your software can decide how far the singer sits from the target. Second, pitch shifting moves the voice slightly in real time. The hard part is doing this gently, so vibrato stays natural and the correction sounds musical instead of robotic.

Why This Is a Good Topic

This is a strong science fair topic because you can measure both signal quality and human response. You can test how well the plugin corrects quartertone targets, how much vibrato survives, and whether listeners familiar with Arabic music prefer one version over another. The project connects to a real artistic problem, since many digital tools do not support non-Western tuning systems well. You can also learn audio DSP, machine learning pitch tracking, user testing, and basic statistics.

Research Questions

• How does the pitch-correction strength affect how accurately singers land on maqam quartertones?
• What is the effect of vibrato rate on the plugin's ability to preserve expressive pitch movement?
• Does a CREPE-based tracker produce cleaner real-time correction than a simpler pitch detector for maqam vocals?
• To what extent do listeners familiar with Arabic music prefer corrected vocals over uncorrected vocals?
• Which correction settings best balance quartertone accuracy and natural vocal tone?
• How does background accompaniment complexity change the plugin's pitch-tracking accuracy?

Basic Materials

• Laptop or desktop computer with audio interface support.
• Digital audio workstation that supports VST or audio units.
• Condenser microphone or quality dynamic microphone.
• Audio interface with low-latency monitoring.
• Headphones for closed-back listening.
• Test vocal recordings or volunteer singer access.
• Reference recordings of maqam scales or phrases.
• Spreadsheet software for listener ratings and pitch error tables.
• Measurement script or plugin logs for pitch output and correction amount.

Advanced Materials

• Computer with GPU support for model testing.
• JUCE or FAUST development environment.
• CREPE model implementation or equivalent pitch tracker.
• Python with NumPy, SciPy, Pandas, and Librosa.
• Praat or similar acoustic analysis software.
• MATLAB or R for signal analysis and statistics.
• Audio annotation tool for frame-by-frame pitch marking.
• Survey platform or custom listening test interface.
• High-quality interface and calibrated monitoring setup.
• Ethics approval materials for human listener testing, if needed.

Software & Tools

• JUCE: Builds the real-time audio plugin and handles cross-platform audio input and output.
• FAUST: Lets you prototype signal processing blocks before turning them into a plugin.
• Python: Processes pitch tracks, computes error metrics, and organizes listener data.
• Librosa: Extracts audio features and helps compare corrected and uncorrected recordings.
• Praat: Measures pitch contours and vibrato features in recorded vocals.

Experiment Steps

1. Define the exact maqam targets you want the plugin to hit and decide how you will represent quartertones numerically.
2. Choose one pitch-tracking method as your baseline and plan a second version for comparison.
3. Design a correction rule that moves pitch toward the target while leaving short-term vibrato movement intact.
4. Plan a test set of vocal phrases that includes stable notes, ornamented notes, and different accompaniment conditions.
5. Build your evaluation plan so you can score pitch accuracy, vibrato preservation, and listener preference with the same recordings.
6. Decide how you will analyze agreement between listener ratings and the acoustic measurements.

Common Pitfalls

• Using a pitch tracker that latches onto harmonics instead of the singer's true fundamental, which makes the correction jump to the wrong note.
• Correcting pitch too aggressively, which flattens vibrato and makes the voice sound synthetic.
• Testing only on isolated sustained notes, which misses the ornamentation and slides that matter in maqam singing.
• Forgetting to separate listener familiarity with Arabic music, which can blur preference results.
• Measuring success only by average pitch error, which can hide ugly short bursts of tracking instability.

What Makes This Competitive

A competitive version needs more than a working plugin. You should compare at least two tracking or correction strategies, then show clear evidence about accuracy, latency, and musical feel. Strong projects also test real listeners with a clean A/B design and analyze the results with proper statistics. If you can connect the signal metrics to what trained ears hear, your project gets much stronger.

Project Variations

• Test the plugin on a different tuning system, such as Persian or Turkish microtonal scales, and compare how well the same correction logic adapts.
• Use solo violin or oud instead of voice, then measure whether the tracker handles continuous pitch differently from vocal vibrato.
• Compare CREPE with a classical pitch detector and evaluate which one better preserves expressive ornaments in real time.

Learn More

• MIT OpenCourseWare, 6.003 Signals and Systems: Learn the signal-processing ideas behind pitch tracking and correction by searching MIT OpenCourseWare.
• JUCE Documentation: Read the free plugin development guides and audio engine references on the official JUCE site.
• FAUST Documentation: Study the language for audio signal processing on the official FAUST site.
• PubMed: Search for review articles on pitch perception, vibrato, and music cognition to ground your listener study.
• IEEE Xplore or arXiv: Search for papers on CREPE, pitch tracking, and real-time vocal processing, then use abstracts and open-access PDFs when available.
• US National Library of Medicine and NIH resources: Look for articles on auditory perception and speech or music processing through PubMed and NIH pages.