Breathing Pitch Waves — User Guide

Pitch-contour modulation for voiced material using a compound breathing-wave controller, stochastic flutter and gasp components, an increasing drive envelope, and bounded semitone displacement.

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 Quick start Presets Parameters Modulation controller Processing pipeline Channels & output Visualization

What this does

Breathing Pitch Waves analyzes the source pitch contour, builds a new target contour at a fixed 100 Hz control rate, and resynthesizes the sound with Praat Manipulation. The generated semitone displacement combines several periodic components with stochastic amplitude variation, then passes the sum through a tanh limiter so the requested pitch depth is a true maximum displacement.

The generated displacement is applied to the detected source F0 at each control time. The original melody or intonation therefore remains the reference contour; the script does not replace the entire file with a contour around one median pitch.

Input scope: The effect requires usable periodic/voiced material. Pitch analysis is performed with Praat To Pitch. If no usable pitch is detected, the script stops instead of creating an arbitrary target pitch.

Random variation: The flutter and gasp terms call randomUniform() while the control curve is built. There is no random-seed parameter, so repeated runs with the same settings can produce different modulation details.

Quick start

Select exactly one Sound object.
Run Breathing_Pitch_Waves.praat.
Choose Manual or one of the seven built-in presets.
For Manual mode, set Breath_rate, Pitch_depth_semitones, Micro_flutter, and Emotional_intensity.
Set Time_step, Minimum_pitch, and Maximum_pitch for pitch analysis.
Choose the Output_sample_rate and whether to draw the visualization and play the result.
Click OK. The final Sound is named <source>_breathing.

Presets

Presets override only the four modulation controls shown below. Pitch-analysis settings, output sample rate, visualization, and playback remain at the values entered in the form.

Preset	Breath rate	Depth	Micro flutter	Emotional intensity
Gentle Breath	0.20 Hz	8 st	1.0	1.2
Emotional Swell	0.25 Hz	15 st	2.0	2.0
Dramatic Breath	0.35 Hz	24 st	5.0	3.0
Panic Breathing	0.80 Hz	36 st	8.0	4.0
Subtle Tremor	0.15 Hz	6 st	3.0	0.8
Deep Meditation	0.10 Hz	4 st	0.5	0.5
Intense Gasping	0.50 Hz	30 st	6.0	3.5

Parameters

Modulation

Parameter	Default	Behavior
Preset	Manual	Manual plus seven built-in modulation presets.
Breath_rate	0.3 Hz	Base phase rate in cycles per second. It also scales the phase rates of the flutter, tremor, and gasp components.
Pitch_depth_semitones	18 st	Maximum absolute semitone displacement. Values are clamped internally to 0–48.
Micro_flutter	4	Scales the two high-rate stochastic flutter components. Values are clamped to 0–20.
Emotional_intensity	2.5	Controls the time-varying drive applied before `tanh`. Values are clamped to 0–10.

Pitch analysis

Parameter	Default	Behavior
Time_step	0.005 s	Time step passed to Praat `To Pitch`. This is separate from the fixed 10 ms modulation-control step.
Minimum_pitch	50 Hz	Lower bound for pitch analysis and for the generated target F0.
Maximum_pitch	900 Hz	Upper bound for pitch analysis and for the generated target F0. It must be greater than `Minimum_pitch`.

Output

Parameter	Default	Behavior
Output_sample_rate	44100 Hz	If different from the source rate, the completed result is resampled with precision 50.
Draw_visualization	Yes	Draw the source/result waveforms, modulation curve, component description, run statistics, and summary strip.
Play_result	Yes	Play the final Sound after processing.

Modulation controller

The controller is evaluated every 10 ms. For each control time t:

phase = (t - xmin) × 2π × Breath_rate breath_fundamental = sin(phase)^3 breath_harmonic = 0.6 × sin(2 × phase)^5 breath_subharmonic = 0.3 × (sin(0.5 × phase)^2 - 0.5) breath_curve = fundamental + harmonic + subharmonic flutter = Micro_flutter × 0.15 × [sin(12 × phase) × U(0.6,1.4) + 0.5 × sin(23.7 × phase) × U(0.8,1.2)] tremor = 0.8 × sin(7.3 × phase) × cos(2.1 × phase) gasp = 3 × [sin(3 × phase)^8 - 35/128] × U(0.5,1.5) emotional_drive = 1 + Emotional_intensity × time_factor^1.5 bounded_motion = tanh((breath_curve + flutter + tremor + gasp) × emotional_drive) total_shift = Pitch_depth_semitones × bounded_motion

Because tanh() stays within −1…+1, Pitch_depth_semitones bounds the requested displacement. The subsequent F0 clamp can reduce the realized displacement when the shifted target would fall outside the selected pitch range.

Processing pipeline

Pitch-analysis reference: a multichannel source is converted to mono for analysis only; a mono source is copied. To Pitch uses the form's time step and pitch bounds.
Detection check: the median detected F0 is queried only to verify/report usable pitch. It is not used as the synthesis reference.
Control grid: the modulation contour is evaluated at a fixed 0.01 s step across the full duration. The number of control points is ceiling(duration / 0.01) + 1, with a minimum of two points.
Source-contour modulation: at each control time, the detected source F0 is read by linear interpolation and multiplied by 2^(total_shift/12). The target is then clipped to the minimum/maximum pitch range.
Per-channel resynthesis: each original channel receives its own Manipulation object. The same generated target PitchTier is inserted into every channel, and each channel is resynthesized with Get resynthesis (overlap-add).
Rebuild: the resynthesized channels are written back into an N-channel Sound with the original time domain and original sampling rate.
Optional resampling: if requested, the complete result is resampled to Output_sample_rate with precision 50.
Peak safety: if the final absolute peak exceeds 0.95, Scale peak: 0.95 attenuates it. Quieter results are not boosted.

Channels & output

Output name: <source>_breathing.
Channel count: preserved. The mono conversion is used only to obtain one shared pitch-analysis reference.
Channel processing: channels are resynthesized independently, but all receive the same target pitch contour.
Duration: processing modifies pitch only, so the output keeps the source time domain.
Sample rate: equals Output_sample_rate after optional resampling; otherwise it remains the original rate.
Level handling: attenuation-only peak safety at 0.95; quieter results retain their level.
Intermediate objects: analysis, Manipulation, channel-work, resynthesis, and PitchTier objects are removed automatically.

Visualization

When Draw_visualization is enabled, the script draws one Picture page with:

Original waveform.
Breathing result waveform.
Pitch-shift control curve in semitones, with a zero reference line.
Component strip describing the controller stages.
Run statistics for rate, depth, flutter, and intensity.
Summary strip identifying the transformation and directing detailed run parameters to the Info window.

Display scaling: the original and result waveform panels are auto-scaled independently, so their plotted heights should not be used as a direct amplitude comparison. The modulation curve is display-decimated deterministically to at most 500 stored points; the DSP uses the full 100 Hz control grid.

Notes & troubleshooting

No usable pitch detected: choose material with voiced/periodic content and set Minimum_pitch and Maximum_pitch so they bracket the source F0. The script stops if the analysis contains no usable pitch.

Requested depth seems smaller: the semitone controller is bounded by the requested depth, but the final target F0 is also clipped to the analysis pitch range. A narrow pitch range can therefore limit the realized shift.

Different result on another run: stochastic amplitudes are generated for flutter and gasp at control points, and no random seed is exposed in the form.