Adaptive Pitch Shifter – User Guide

What this does

The script modifies the pitch tier of a Praat Manipulation object. At each voiced pitch-tier point, it derives a bipolar modulation signal, optionally smooths that signal, combines it with a base pitch multiplier, writes the resulting frequency back to the pitch tier, and resynthesizes the sound.

The modulation source can follow the sound's measured intensity, its detected pitch contour, a sine-wave LFO, or the average of amplitude modulation and the LFO. The source-dependent modulation is centred around zero, so Base_pitch_shift acts as the centre of the pitch multiplier.

Quick start

1. Select exactly one Sound object in Praat.
2. Run Adaptive_Pitch_Shifter.praat.
3. Choose Custom or one of the six named presets.
4. Set the processing, quality, visualization, and playback options.
5. Click OK. The output Sound remains selected when processing finishes.

Presets

The menu contains Custom plus six named presets. A named preset overwrites only Base_pitch_shift, Modulation_amount, Modulation_source, LFO_frequency, and Smoothing_factor. The other form settings remain as entered by the user.

LFO_frequency is used only by Time-based LFO and Combined. A stored LFO value therefore has no effect when the selected source is Amplitude or Pitch Contour.

Parameters

Pitch and modulation controls

Processing and output controls

Quality

Quality directly sets the analysis time step and pitch bounds used by the Manipulation analysis. The same time step and pitch floor are also used by the amplitude-envelope analysis; Pitch Contour additionally uses the corresponding pitch ceiling.

Modulation

The script computes a raw bipolar modulation value m at each voiced pitch-tier point.

Amplitude

Praat Intensity is measured across the input using the current quality time step and pitch floor. The script takes the minimum and maximum measured Intensity values over the sound and maps the value at each voiced point to approximately −1…+1:

normalized = clamp((Intensity(t) - Intensity_min) /
                   (Intensity_max - Intensity_min), 0, 1)
m_raw = 2 * normalized - 1

The Intensity object is in Praat's dB SPL scale. If the measured range collapses to a single value, the modulation at that point remains zero.

Pitch Contour

A separate Pitch analysis is measured over the input. Its detected minimum and maximum frequencies define the adaptive range:

normalized = clamp((Pitch(t) - Pitch_min) /
                   (Pitch_max - Pitch_min), 0, 1)
m_raw = 2 * normalized - 1

If the detected pitch range collapses to a single value, the modulation at that point remains zero.

Time-based LFO

m_raw = sin(2 * pi * LFO_frequency * time)

The sine wave is bipolar, so the base pitch multiplier is its centre.

Combined

Combined modulation averages the bipolar amplitude signal and the sine LFO:

m_raw = (amplitude_modulation + lfo) / 2

Smoothing

The first available modulation point is used directly. Each later voiced point is smoothed with a one-pole recurrence:

m = m_raw * (1 - Smoothing_factor)
    + previous_m * Smoothing_factor

The smoothing is applied to the modulation state before it is combined with the base pitch shift.

Pitch transformation

For each voiced pitch-tier point, the script converts the smoothed modulation into a multiplicative pitch factor:

pitchMultiplier = Base_pitch_shift
                  + m * Modulation_amount
pitchMultiplier = max(0.05, pitchMultiplier)
newFrequency = originalFrequency * pitchMultiplier
newFrequency = clamp(newFrequency, 20, 2000) Hz

The 0.05 multiplier floor prevents a non-positive multiplier under strong negative modulation. The final 20–2000 Hz bounds are safety limits on written pitch-tier frequencies; they do not change the analysis bounds selected by Quality.

Resynthesis & stereo

Resynthesis path

The script changes the pitch tier only; it does not add or modify a DurationTier.

Stereo width

When Add_stereo_width is enabled and the resynthesized Sound is mono, the mono result is duplicated to two channels. The left channel remains unchanged. The right channel becomes:

right = 0.85 * current
        + 0.15 * left delayed by 12 ms

The delay is converted to samples with the input sampling frequency. This widening stage is applied before Output_gain.

Visualization

When Draw_visualization is enabled, the script samples pitch/modulation points across the entire file for display. At most about 500 points are retained, using a regular stride through the complete pitch tier; this decimation affects only the visualization, not the pitch transformation.

The Picture output contains:

• Input waveform.
• Shifted waveform; for multichannel output, channels 1 and 2 are drawn separately.
• Pitch contour comparison on one shared frequency scale: original in grey and shifted in blue.
• Modulation signal using the actual displayed modulation range with a zero reference line.
• Output spectrogram; channel 1 is used when the output has more than one channel, and the upper frequency is min(5000 Hz, Nyquist).
• Summary strip with preset, base shift, source, amount, smoothing, pitch-point count, resynthesis path, stereo-width state, and LFO frequency when applicable.

The input and output waveform panels are drawn with their own automatic amplitude ranges, so their plotted heights are not a shared amplitude comparison.

Outputs

• Output object: one resynthesized Sound, left selected at the end of the script.
• Name pattern: <source>_shifted_<preset-name>.
• Preset suffixes: Custom, Subtle Wobble, Robot Voice, Harmonic Shimmer, Deep Bass, Vibrato, or Extreme Warp.
• Duration: pitch is modified without changing the DurationTier.
• Final level: the samples are multiplied by Output_gain; there is no subsequent normalization.
• Playback: controlled by Play_result.
• Picture output: controlled by Draw_visualization.

The Info window reports the source, selected preset and modulation source, base shift, modulation amount, LFO frequency when applicable, creation progress, selected resynthesis method, output name, and final gain.