Algorithmic Metallic Synthesis — User Guide

What this does

Algorithmic Metallic Synthesis creates a new sound from scratch; no input Sound object is required. Each synthesis voice consists of a regular strike train convolved with an exponentially decaying FM ring kernel. The kernel supplies an inharmonic carrier, a decaying modulation index, and an amplitude decay, while the strike train determines when that resonance is excited.

Because the complete ring kernel is reproduced at every strike, resonance tails can overlap naturally when strikes occur faster than the decay. Multiple voices and layers are summed, then the result receives a global edge fade, optional stereo rendering, optional target-peak normalization, visualization, and playback.

Quick start

1. In Praat, run Run script… → Algorithmic Metallic Synthesis.praat. No sound selection is needed.
2. Choose a Preset, or keep Custom to use the settings below the preset menu.
3. Set the synthesis and output controls as needed.
4. Click OK. The generated object is named metallic_<preset>.

For a clear first example, try Temple Bell for separated, long-decay resonances or Gamelan for a denser multi-voice texture.

Synthesis mechanism

Strike train × ring kernel

For each voice, the script creates a unit pulse train on exact sample positions. The nominal strike rate is converted to an integer period in samples, so the realized rate is sample_rate / period_samples. With randomization enabled, each voice also receives a random initial offset within one period.

The pulse train is convolved with the voice's ring kernel:

h(t) = A · exp(-t / τ) ·
       sin(2πfc t + β · exp(-t / τM) · sin(2πfm t))

τ  = realized resonance decay
τM = 0.60 · τ
fc = carrier frequency after detuning and anti-alias adjustment
fm = layer modulation rate
β  = modulation depth

The kernel normally lasts five decay constants, with a 30 ms minimum and a cap at the requested output duration. Its modulation index decays faster than its amplitude, so each ring tends to move from a more strongly modulated onset toward a simpler oscillation as it decays.

Overlap and level compensation

Strike tails are allowed to overlap. To prevent strike density from becoming an unintended gain control, each voice is attenuated when its average overlap estimate strike_rate × decay exceeds one. Voice amplitude is also scaled by the number of layers and by the square root of the number of voices in that layer.

Parameters

Presets

Preset selection occurs before validation. Fields not listed for a preset retain the values currently entered in the form. In particular, presets do not change Sample_rate_Hz, Random_seed, Normalize_output, Draw_visualization, or Play_result.

Synthesis modes

The Chaotic Resonance label describes the mode's irregular behavior. Its control law is stochastic when randomization is enabled and a fixed irregular lattice when disabled, rather than a named deterministic-chaos equation.

Spatial modes

Both stereo modes are created from the synthesized mono signal; there is no pre-existing spatial image to preserve.

Randomization & seed

With Randomize_parameters = Yes, the script randomizes layer modulation rate and decay, voice frequency and modulation depth, and the initial strike offset. Chaotic Resonance additionally randomizes its carrier construction, detuning, modulation depth, and strike rate.

A positive Random_seed reproduces these randomized choices. Seed 0 uses an unpredictable stream. With Randomize_parameters = No, the synthesis parameters and strike offsets are deterministic, so the audible result does not depend on the seed.

Visualization

When Draw_visualization is enabled, the script draws a model/measurement view from the actual values used by the DSP.

A — Measured output waveform

Shows the final output waveform on a symmetric amplitude scale. For stereo output, the channel with the higher RMS is displayed; the channels are not folded to mono.

B — Actual strike-resonance field

Plots carrier frequency against time. Rust vertical ticks mark realized strikes; pale horizontal lines show each strike's actual ring-kernel duration. For large event counts, the drawing is subsampled for legibility while the synthesis itself remains unchanged.

C — Model → measurement

Paints a measured spectrogram of the same representative output channel and overlays horizontal guides at the actual realized carrier frequencies. When many voices are present, carrier guides are subsampled for legibility.

The lower mechanism strip summarizes the pulse-train/convolution model. The final dashboard reports realized carrier range, mean decay, strike count, anti-alias adjustments, skipped voices, output peak/RMS, spatial mode, and seed.

Output & safeguards

• Object name: metallic_<preset>; Custom produces metallic_Custom.
• Duration: exactly the requested output duration after any preset override. Convolution tails beyond the endpoint are cropped.
• Sample rate: the selected Sample_rate_Hz.
• Normalization: when enabled and the signal is nonzero, Scale peak: 0.9 performs target peak normalization. When disabled, the script reports a warning if the peak exceeds 0.99.
• Anti-alias guard: carriers are kept below 0.45 × sample_rate with additional margin for the estimated FM sideband spread. A carrier may be moved downward; a voice is skipped if no safe carrier remains.
• Workload guards: synthesis is rejected above 160 estimated oscillator voices or above the script's 100,000-strike workload estimate.

Compositional context

Metallic timbre synthesis has a long computer-music lineage built around inharmonic frequency relationships, time-varying spectra, and decaying resonances. This tool approaches that territory through a specific algorithmic design: regularly triggered pulse trains convolved with decaying FM ring kernels, allowing the same resonance model to range from isolated bell-like strikes to dense overlapping metallic fields.