lesson
Updated 6 days ago ยท 1 view
Why can auto-tune lock an out-of-tune solo vocal onto pitch in milliseconds, but turns into a garbled, metallic mess the moment two people sing at once?
The secret lies in how audio software tracks the physical repetition of a single vibrating source.
What is the Fundamental Frequency (f0โ)?
When a vocalist sings a note, vocal folds vibrate at a primary repeating rate called the fundamental frequency (f0โ).
Even though a vocal waveform looks complex due to upper harmonics, the entire pattern repeats over a single duration called the fundamental period (T), where f0โ=T1โ.
๐Interactive audio waveform and harmonic spectrum diagram. Top card shows a complex vocal waveform over time with period markers (T) highlighting the repeating cycle duration. A toggle button switches between 'Time Domain (Waveform)' and 'Frequency Domain (Harmonics)'. When in Frequency view, show a bar graph of frequencies with tall f0 (e.g., 220 Hz) and integer harmonic multiples (440 Hz, 660 Hz, 880 Hz) clearly labeled. Clean light UI (#ffffff card, #1e2945 text, #0284c7 blue accents, responsive to 350px width).
How does a digital audio workstation detect this exact period T from millions of raw sample values every second?
Time-Domain Detection: Autocorrelation
Autocorrelation is a mathematical tool that compares a signal with a time-delayed copy of itself to find repeating patterns.
In 1948, mathematician Norbert Wiener formalized autocorrelation for signal analysis, defining the correlation R(ฯ) at a time delay (lag) ฯ as: R(ฯ)=โnโx[n]โ
x[nโฯ]
When the lag ฯ exactly equals the true wave period T, the delayed peaks align with the original peaks, producing a huge spike in R(ฯ).
๐Animated visual explaining autocorrelation. Top graph shows the original waveform x[n] in solid navy (#1e2945) and a shifted waveform x[n - tau] in dashed blue (#0284c7). A slider or auto-play sweeps the lag tau from 0 to 10 ms. Bottom graph plots R(tau) against tau, showing a massive initial peak at tau = 0, followed by a distinct second peak at tau = T (e.g., 4.54 ms for 220 Hz). A callout box dynamically calculates f0 = 1000 / 4.54 ms = 220.2 Hz. Light theme, card container.
What happens when strong formant resonances or acoustic reflections create false peaks in the time waveform?