Pitch Stylization and Shift — User Guide

Pitch-contour stylization, chromatic quantization, monotone transformation, transposition, and optional formant following with duration-preserving Praat resynthesis.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.6.1 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools
Contents:

What this does

Pitch Stylization and Shift analyzes the F0 contour of one Sound and builds a processed PitchTier. Depending on the selected preset, it can simplify the contour, quantize voiced pitch points to the nearest equal-tempered semitone, flatten the contour to a single register, transpose the result, or combine stylization and transposition.

The processed PitchTier is applied independently to every original channel with Praat Manipulation and overlap-add resynthesis. An optional formant-follow stage can then move the spectral envelope in the same general register direction while leaving pitch, pitch range, and duration multipliers at 1 inside Praat's Change speaker operation.

Important: stylization is used only in Normal mode. Quantize and Robot build their own target contours and do not use Stylize_Hz.

Quick start

  1. Select exactly one Sound object in Praat.
  2. Run Pitch_Stylization_and_Shift.praat.
  3. Choose a preset, or choose Manual and set Stylize_Hz and Shift_semitones.
  4. Set Formant_follow_percent if you want timbre/formants to follow the measured register change.
  5. Set the pitch-analysis range and time step for the source material.
  6. Choose whether to draw the visualization and play the result.
Analysis range: Min_pitch and Max_pitch control pitch detection. They are not the final synthesis limits. Processed targets are separately constrained to 20 Hz through 0.45 × sample rate.

Presets

The menu contains six built-in transformations plus Manual. A preset overrides the internal stylize amount, shift amount, and processing mode shown below. It does not override Formant_follow_percent, the analysis settings, or the visualization/play switches.

PresetModeStylizeShiftResult
ManualNormalUser valueUser valueDirect manual stylization and/or transposition.
Gentle SmoothingNormal3 Hz0 stSimplifies the detected pitch contour without intentional register shift.
Stepwise Quantize (Auto-tune)Quantize—0 stMoves each voiced tier point to the nearest 12-TET semitone referenced to A4 = 440 Hz.
Robot Voice (Monotone)Robot—−2 stFlattens the contour to its geometric-mean F0, then lowers that centre by 2 semitones.
Pitch Down (−1 Octave)Normal0 Hz−12 stTransposes the original detected contour down one octave.
Pitch Up (+1 Octave)Normal0 Hz+12 stTransposes the original detected contour up one octave.
Strong StylizeNormal10 Hz−5 stStrong contour simplification followed by a five-semitone downward transposition.

Pitch processing

Normal mode

The source PitchTier is copied. If Stylize_Hz > 0, Praat's Stylize: …, "Hz" simplifies that working tier. Each resulting point is then optionally multiplied by the transposition ratio:

targetF0 = stylizedF0 × 2^(Shift_semitones / 12)

Quantize mode

Each detected voiced point is quantized independently to the nearest equal-tempered semitone around A4 = 440 Hz:

targetF0 = 440 × 2^( round(12 × log2(F0 / 440)) / 12 )

The preset itself uses zero additional transposition.

Robot mode

The script computes the geometric mean of the original PitchTier points in log-frequency space. Every processed pitch point is set to that common centre and then shifted by the preset's −2 semitones.

Synthesis safety

After the selected transformation and transposition, each target point is constrained to 20 Hz … 0.45 × source sample rate. The number of limited points is reported in the Info window.

Identity path

Manual + Stylize 0 Hz + Shift 0 st bypasses PSOLA and begins from a direct copy of the source. Pitch analysis still runs for reporting/visualization. The final common peak-safety stage is still active, so a source whose absolute peak exceeds 0.95 will be attenuated to 0.95; otherwise the copied audio is unchanged.

If no usable voiced pitch is detected, this identity setting can still return the copied Sound. Any non-identity transformation stops with a clear message when no usable pitch is available.

Formant follow

Formant_follow_percent couples timbral/formant movement to the measured mean pitch-register change of the processed PitchTier. The mean change is measured in semitones at the original PitchTier time points.

formantRatioRaw = 2^((meanPitchChange_st × Formant_follow_percent / 100) / 12)

The formant ratio is safety-limited to 0.5 … 2.0. Formant processing runs only when the follow amount is above zero and the resulting ratio differs from 1.

Praat Change speaker is applied separately to every output channel with:

formant multiplier = formantRatio
pitch multiplier = 1
pitch-range multiplier = 1
duration multiplier = 1

The pitch-analysis range used by Change speaker is adapted to the measured processed register and capped below 0.45 × sample rate. If the requested ratio was clipped to the 0.5–2.0 safety range, the raw ratio is reported in the Info window.

Parameters

ParameterDefaultMeaning
PresetManualManual plus six built-in pitch transformations.
Stylize_Hz2 HzPraat PitchTier stylization frequency in Normal mode; must be ≥ 0.
Shift_semitones0 stTransposition applied after Normal/Quantize/Robot target generation; validated to −48 … +48 st.
Formant_follow_percent0%0 preserves the no-formant-follow path; 100% maps the measured mean register change directly to the formant ratio before safety limiting.
Time_step0.005 sPitch-analysis and Manipulation time step.
Min_pitch75 HzLower analysis bound.
Max_pitch600 HzUpper analysis bound; must be below 45% of the source sampling frequency.
Draw_visualizationYesDraw the current run in the Picture window.
Play_resultYesPlay the final Sound.

Output behavior

Visualization

When Draw_visualization is enabled, the Picture window contains:

For an identity run on material with no detected voiced pitch, the pitch panel explicitly reports that no voiced pitch was detected.