Editing Basics

Editing here means changing the sound of one part: its pitch, its length, or the character of a voice.

It is different from analysis, which only measures a track and leaves it untouched. It is also different from mixing, which balances several parts together.

This page defines the words you need before reading the Editing DSP guide. A short example near the end shows the two main edits as calls.

Where editing sits

You usually edit individual parts (fix timing, tune a note, change a voice), then mix them together, then master the finished stereo file for delivery. Editing is the per-part repair-and-shape step. Conservative edits keep the part natural; aggressive edits are a creative effect.

Pitch and time are independent

Speed up a tape and the music gets both faster and higher — the two are coupled. The whole point of modern editing DSP is to separate them, so you can change one without dragging the other along:

• Time stretch changes the duration without changing the pitch. Here rate is a playback-speed multiplier: a rate above 1.0 plays faster, so the clip gets shorter; below 1.0 plays slower, so it gets longer.
• Pitch shift changes the pitch without changing the duration.

Doing this cleanly is hard: the algorithm has to invent or remove time while keeping each note's frequency, so very large moves always leave some artifacts (smearing, a "phasey" or robotic tone). Small moves stay transparent.

PARAM SWEEP · TIME STRETCHIDLE

Time stretch — changing length, not pitch

Scroll to load

Time stretching changes duration while keeping pitch approximately unchanged. A rate below 1 slows and lengthens the clip; a rate above 1 speeds and shortens it. The waveform shows the duration change. Each render is peak-normalized for headroom, which does not equalize perceived loudness. Listen for changes in timing and transient clarity.

Preview

Switching during playback restarts the selected audio from the beginning.Processed

Rate

1 ×

PARAM SWEEP · PITCH SHIFTIDLE

Pitch shift — moving the whole harmonic comb

Scroll to load

Pitch shifting transposes a sound without changing its length. Drag the semitones and watch the spectrum: every harmonic scales together and the fundamental marker tracks the new pitch. Because this shift is not formant-preserving, the formant bumps move too — the "chipmunk" effect. Press play to hear it.

Preview

Switching during playback restarts the selected audio from the beginning.Processed

Pitch

0 st

Semitones and cents

Pitch in music is measured in semitones, the smallest step on a piano keyboard. Twelve semitones make one octave, and an octave doubles the frequency: A4 = 440 Hz, A5 = 880 Hz.

So semitones = 12 shifts up one octave, -12 shifts down one octave, and 7 is a perfect fifth.

A cent is 1/100 of a semitone. It is used for fine tuning; most listeners start to notice pitch error past roughly ±5-10 cents.

MIDI note numbers

When a function asks for a MIDI note number (such as pitchCorrectToMidi's currentMidi and targetMidi), it wants an integer (or fractional) index where each whole number is one semitone. Two anchors are worth memorizing:

• A4 = 69 = 440 Hz — the tuning reference.
• C4 = 60 — "middle C".

Every octave adds 12, so C5 = 72, C3 = 48. A few common targets:

Note	MIDI	Frequency
C3	48	130.81 Hz
C4 (middle C)	60	261.63 Hz
E4	64	329.63 Hz
G4	67	392.00 Hz
A4 (tuning reference)	69	440.00 Hz
C5	72	523.25 Hz

To convert by hand: midi = 69 + 12 · log2(freq / 440) and freq = 440 · 2^((midi − 69) / 12). A fractional MIDI value like 68.7 is simply a pitch a little below A4 — useful because a singer is rarely exactly on a note, and you can pass the measured pitch as the current value and a whole number as the target.

Formant: pitch's independent partner

A formant is a resonance peak in the spectrum's envelope, set by the shape of the vocal tract. It moves when the vowel changes, not when the pitch does. In instruments, similar resonances come from the body of the instrument.

Formants shape vowels and the perceived size or character of a voice. They are different from pitch: a singer can sing different notes while the vowel character still comes from the same vocal-tract resonances.

This is why voice tools keep pitch and formant on separate controls.

Control	What changes	What you hear
Pitch	The musical note	The same voice singing higher or lower
Formant	The vocal character	A smaller/brighter or larger/darker voice at the same note

Moving both together too aggressively creates the familiar artificial "chipmunk" effect. By default in voiceChange (formantMode: 'relative') a pitch shift moves the formants too, so formantFactor: 1 with +12 semitones sounds like a chipmunk. Set formantMode: 'absolute' with formantFactor: 1 to keep the voice's size while changing the note: the factor is then measured against the input, so 1 leaves the formants where they were whatever the pitch shift.

Samples vs seconds

Editing functions that act on a region (like note stretch) take sample offsets, because sample positions are exact and never drift. Most also accept the same position in seconds under a *Sec name (onsetSec, offsetSec, startSec, endSec, paddingSec), converted at the call's sample rate; give one spelling or the other, never both. The conversion is samples = round(seconds × sampleRate) — at 48 kHz, 0.25 s is sample 12000. See Audio Basics for what a sample and sample rate are.

Try it in libsonare

Time stretch and pitch shift are one call each. A rate of 0.5 plays at half speed, so the result is twice as long at the same pitch; semitones: 12 raises the pitch an octave and keeps the length.

Browser

import { init, timeStretch, pitchShift } from '@libraz/libsonare';

await init();

const slower = timeStretch(samples, sampleRate, 0.5);   // twice as long, same pitch
const higher = pitchShift(samples, sampleRate, 12);     // one octave up, same length

Python

import libsonare as sonare

slower = sonare.time_stretch(samples, sample_rate, 0.5)   # twice as long, same pitch
higher = sonare.pitch_shift(samples, sample_rate, 12)     # one octave up, same length

CLI

sonare time-stretch take.wav --rate 0.5 -o slower.wav
sonare pitch-shift take.wav --semitones 12 -o higher.wav

How libsonare implements editing DSP

The implementation uses the same C++ DSP core as analysis and mastering, but these functions rewrite the signal instead of measuring it.

Function family	Implementation idea	Beginner takeaway
Time stretch / pitch shift	Phase vocoder (rebuilds the signal from short overlapping spectral frames) plus resampling	Duration and pitch can be changed separately, but large moves create artifacts
voiceChange	Shift the spectral envelope separately from harmonic pitch	Voice character can move without changing the target note
pitchCorrectToMidi	Move from a caller-supplied current MIDI note to a target MIDI note	Estimate the current pitch first with pitchYin, pitchPyin, or your own detector
autoTune	Snap a monophonic recording to the notes of a detected (key: 'detect') or named key, with a strength control	Returns the corrected samples and the key it used
noteStretch	Process an exact sample-offset region	Pass onsetSample / offsetSample, or onsetSec / offsetSec

Related: Editing DSP, Audio Basics, MIR Overview, JavaScript API