Dynamic Formant Sweeper — User Guide

What this does

Dynamic Formant Sweeper creates a moving resonance-like spectral emphasis controlled by an LFO. It first measures a robust F1 landmark from the selected sound using Praat's FormantPath analysis. That landmark is used only as a reference region: the script does not resynthesize the sound with LPC, does not inverse-filter an excitation signal, and does not use a FormantGrid.

Instead, the sound is processed in short overlapping spectral grains. For each grain, the script reduces energy around the measured F1 region and adds a broad spectral-envelope peak at the current LFO target frequency. The same real-valued gain is applied to the real and imaginary parts of every spectral bin, so the original phase is preserved bin by bin.

Quick start

1. Select exactly one Sound object.
2. Choose a preset or Manual.
3. Set the LFO rate, minimum and maximum target frequencies, and waveform.
4. Use Envelope_width_Hz for a broader or narrower target region and Sweep_strength_dB for effect depth.
5. Adjust Dry_wet_mix. A value of 0 returns a bypass copy without running the FormantPath or STFT processing.
6. Leave Preserve_frame_energy on for more stable grain-to-grain level unless you specifically want the spectral reshaping to alter local energy more freely.

Signal flow

F1 landmark

The default multichannel analysis source is the loudest channel. The advanced dialog can instead use channel 1 or a mono sum. FormantPath then supplies the first-formant track, from which the script takes the median, 25th percentile and 75th percentile. The source spectral region is widened according to the F1 interquartile spread so natural F1 motion is not treated as a single narrow frequency.

Adaptive WOLA geometry

The update rate is adaptive: at least 20 updates per second, up to 80, with a target of about 16 updates per LFO cycle. Grains use an adaptive duration between roughly 40 and 100 ms. Hann analysis and synthesis windows are combined with a Hann² weight buffer for normalized overlap-add reconstruction.

Frame-energy preservation

When Preserve_frame_energy is enabled, the script measures spectral energy before and after shaping and applies a compensating correction. That correction is limited to ±3 dB per grain, preventing the energy normalizer itself from creating extreme gain changes.

Presets

Presets replace the LFO rate, target range, envelope width, sweep strength, LFO shape and dry/wet mix. They do not override the energy-preservation switch, optional high-cut, visualization/playback choices, or advanced output settings.

Main controls

Advanced settings

Enable Advanced_settings in the main form to open a second dialog.

Output & visualization

Channels, timing and naming

Every input channel is processed independently with the same measured F1 landmark and the same LFO trajectory. Channel count, sample rate and duration are preserved. Processing is performed on a work copy starting at time 0, and the original start time is restored afterward. The normal output name is <original>_swept_<preset>; the 0% bypass path uses <original>_swept_Bypass. The original Sound remains unchanged.

Output level modes

Visualization

The Picture page is 8 inches high and uses a centered excerpt of at most 8 seconds. The first panel shows the LFO spectral-envelope target and a horizontal line at the measured F1 median. The second panel shows a mono-fold output spectrogram up to 5 kHz or Nyquist, with the target trajectory overlaid in yellow. A summary strip reports the F1 landmark, source/target envelope widths, sweep strength, grain/hop geometry, input/output peak and RMS, and the level action.

Determinism

The processor uses no random numbers or random seed. With the same sound and settings, the LFO trajectory and spectral shaping are deterministic.