Vocal Formant Wavetable Generator

This Python script extracts formant characteristics from a vocal .wav file and generates a wavetable compatible with Ewan Bristow's FreakyTable and Kilohearts Filter Table audio plugins. Using Praat’s Burg method via the parselmouth library, it isolates vowel-shaping resonances for a precise, controlled filter profile.

Functionality

• Analyzes formants in a vocal audio file using Praat’s Burg method. Burg method references: Sound: To Formant (burg)... praat_formant_burg: Formant estimation using the Praat burg algorithm
• Applies a formant envelope to the spectral content of individual audio frames over time to create a wavetable.
• Optimized for clean vocal samples (IE sustained vowels).
• Ensure input .wav sample rate matches the target output (default: 44100 Hz) for accurate formant analysis. Dynamic input file samplerate functionality coming soon.

Comparison to using a raw vocal sample as a wavetable

Unlike using a raw vocal sample as a wavetable, which includes pitch, noise, and articulation etc... this script isolates and emphasizes vowel-shaping resonances. It suppresses pitch and non-formant overtones, producing a clearer, more controlled filter profile for vowel-like effects.

Requirements

• Python 3.6+
• Required libraries: parselmouth, numpy, soundfile

pip install parselmouth numpy soundfile

Key Parameters

Parameter	Effect	Range
--blend	Balances formant envelope and original spectrum	0.0–1.0
--formant-q	Controls width of formant peaks (narrow to wide)	5–30
--formant-strength	Scales magnitude of formant peaks	0.1–3.0+
--contrast-mode	Squashes non-formant content for emphasis	Boolean
--max-frames	Number of wavetable frames	Default: 256
--frame-length	Samples per frame	Default: 2048
--spectral-tilt	High-frequency roll-off	0.0–1.0

Usage

C:\Users\phix\formant_table>python3 formant_table.py --help
usage: formant_table.py [-h] [-o OUTPUT] [--max-frames MAX_FRAMES] [--frame-length FRAME_LENGTH]
                               [--sample-rate SAMPLE_RATE] [--blend BLEND] [--formant-q FORMANT_Q]
                               [--spectral-tilt SPECTRAL_TILT] [--fft-bins FFT_BINS] [--num-formants NUM_FORMANTS]
                               [--formant-strength FORMANT_STRENGTH] [-c]
                               input

Generate formant-isolated wavetable from vocal sample.

positional arguments:
  input                 Input vocal .wav file

options:
  -h, --help            show this help message and exit
  -o OUTPUT, --output OUTPUT
                        Output wavetable .wav file (default: <inputname>-wt.wav)
  --max-frames MAX_FRAMES
                        Number of frames in wavetable (default: 256)
  --frame-length FRAME_LENGTH
                        Samples per frame (default: 2048)
  --sample-rate SAMPLE_RATE
                        Output sample rate (default: 44100)
  --blend BLEND         Blend original spectrum and formants (0–1, default: 0.7)
  --formant-q FORMANT_Q
                        Formant peak sharpness (default: 12.0)
  --spectral-tilt SPECTRAL_TILT
                        High-frequency tilt (0–1, default: 0.4)
  --fft-bins FFT_BINS   FFT bin resolution (default: 512)
  --num-formants NUM_FORMANTS
                        Number of formants to extract (default: 4)
  --formant-strength FORMANT_STRENGTH
                        Scale magnitude of formant envelope (default: 1.0)
  -c, --contrast-mode   Activate contrast mode to emphasize formants by reducing non-formant frequencies.

C:\Users\phix\formant_table>python3 formant_table.py lets_go.wav --formant-strength 2.0
[!]  'lets_go-wt.wav' exists. Overwrite? [y/N]: y

[OK] Generated wavetable: lets_go-wt.wav
   Frames: 256 | Frame size: 2048 | Sample rate: 44100
   Blend: 0.7 | Formant Q: 12.0 | Spectral tilt: 0.4 | Strength: 2.0