Live Pitch Tracker
See the melody of your voice
The same kind of pitch (F0) contour I analyze in my prosody research, drawn live from your microphone.
Note: this is an interactive demo, not the method used in my papers. The papers extract pitch offline with pYAAPT, interpolate unvoiced regions, and align words across languages before measuring correspondence. This tracker uses a simpler real-time algorithm (YIN), so its contours and values will differ from the papers' figures.
Press Start, allow microphone access, and speak or sing.
Things to try
- Ask a question, then make a statement. "You're coming?" rises at the end; "You're coming." falls.
- Stress different words. "I didn't say that" vs. "I didn't say that": watch where the pitch peaks.
- Say a sentence in another language. Tonal languages like Mandarin show pitch shapes on every syllable.
- Notice the gaps. Sounds like "s" and "f" have no pitch, so the line breaks there, just like the unvoiced regions in my papers' figures.
Privacy
Nothing is recorded or kept. Your audio is analyzed in your browser one 40-millisecond slice at a time, and each slice is discarded right after its pitch is measured.
- No audio or pitch data is uploaded, saved to disk, or stored in the browser.
- Only the pitch numbers for the last ~10 seconds are kept in memory to draw the line. Clear, Stop, or leaving the page removes them.
- The microphone turns off when you press Stop, switch tabs, or leave the page.
How it works
The tracker uses the YIN algorithm (de Cheveigné & Kawahara, 2002). About 66 times per second it compares a slice of your voice with time-shifted copies of itself; the shift where the signal best repeats is one pitch period. A confidence threshold separates voiced sounds from silence and consonants, and a short median filter removes stray jumps. The pitch axis is logarithmic, so equal musical steps look equal.
This is a different method from my research pipeline, which uses offline pYAAPT pitch extraction, interpolation through unvoiced regions, word timestamps, and cross-lingual alignment. Use the tracker to explore your own voice, not to reproduce the papers' numbers.