Voice Transformation — User Guide
Combined pitch, duration, formant and band-pass processing with channel-preserving resynthesis.
What this does
Voice Transformation runs up to four stages in a fixed order: source-relative pitch shift, duration scaling, formant transformation with Praat Change gender, and a Hann-band pass filter. Pitch and duration are handled through Manipulation/overlap-add; formant and filter processing follow the resynthesis.
The band-pass filter is always applied. Therefore neutral pitch, formant and duration values are not an identity path unless the chosen filter band itself is effectively transparent for the material.
Quick start
- Select exactly one Sound.
- Choose a preset or Custom.
- For Custom, set the pitch-analysis range, filter band, semitone shift, formant ratio and duration factor.
- Run the script. The result is named
<source>_<presetName>.
Presets
| Preset | Pitch range | Band | Pitch | Formant | Duration |
|---|---|---|---|---|---|
| Vocal Harmonics | 75–500 Hz | 100–4000 Hz | 0 st | 1.0× | 1.0× |
| Reduce Breathiness | 75–500 Hz | 200–3000 Hz | 0 st | 1.0× | 1.0× |
| Deeper Voice | 50–400 Hz | 80–4000 Hz | -4 st | 1.0× | 1.0× |
| Higher Voice | 100–800 Hz | 150–5000 Hz | +5 st | 1.0× | 1.0× |
| Chipmunk Effect | 150–1000 Hz | 200–8000 Hz | +7 st | 1.4× | 0.85× |
| Robot Voice | 50–300 Hz | 100–2500 Hz | -6 st | 0.9× | 1.05× |
| Telephone Effect | 75–500 Hz | 300–3400 Hz | 0 st | 1.0× | 1.0× |
| Radio Voice | 60–400 Hz | 80–8000 Hz | -2 st | 0.95× | 0.98× |
Parameters
| Control | Default | Validation / role |
|---|---|---|
| Pitch_floor / Pitch_ceiling | 75 / 600 Hz | ceiling > floor and below 0.45×SR |
| Time_step | 0.01 s | >0 and ≤0.1; Manipulation pitch analysis |
| Freq_cutoff_low / high | 120 / 3500 Hz | high > low; high below 0.49×SR |
| Pitch_shift_semitones | 0 | -48 .. +48 |
| Formant_shift_ratio | 1.0 | >0 .. 4 |
| Duration_factor | 1.0 | >0 .. 4 |
Processing order
- Create a mono analysis reference. Pitch analysis itself is performed only when Pitch_shift_semitones is nonzero.
- If voiced pitch exists, build a shifted PitchTier with
newF0 = originalF0 * 2^(ST/12), limited to 20 Hz ..0.45*SR. - If Duration_factor differs from 1, build a constant DurationTier over the true xmin/xmax domain.
- For each source channel, create a Manipulation object and replace the shared pitch/duration tiers as applicable.
- Resynthesize with overlap-add.
- If Formant_shift_ratio differs from 1, apply
Change genderwith the requested formant ratio, new pitch median=0, pitch-range factor=1.0 and duration factor=1.0. - Apply
Filter (pass Hann band)using the requested low/high cutoffs and 100 Hz smoothing. - Rebuild the original channel count and apply attenuation-only peak safety at 0.95.
Pitch and channel behavior
Multichannel material is not collapsed for the final audio. A mono reference supplies one shared absolute shifted PitchTier, then every original channel is resynthesized independently and rebuilt into the original channel count. This means all channels receive the same pitch trajectory when pitch shifting is active, while their audio content otherwise remains separate through the channel processing chain.
If no voiced pitch points are detected, the requested pitch shift is skipped cleanly. Duration, formant and filter stages still run.
Visualization
- Original and transformed waveforms are drawn with separate automatic amplitude scales.
- Original and transformed spectrograms are shown up to
min(5000, 0.49*SR). - The original spectrogram marks the requested filter-band boundaries.
- The four summary bars show the requested settings for pitch, formant, duration and band. If pitch detection failed and pitch shifting was skipped, the pitch bar still displays the requested semitone value.
- The stats line reports source/result duration and preset.
Output
- Name suffixes:
_Vocal,_Clean,_Deeper,_Higher,_Chipmunk,_Robot,_Telephone,_Radio, or_Custom. - Channel count: preserved exactly
- Duration: determined by Duration_factor and subsequent formant/filter processing; Info reports the measured final duration
- No explicit resampling stage is used; channel reconstruction uses the processed channels' sampling frequency
- Peak handling: attenuation only above 0.95