Create and review sounds

Write procedural audio in native .ashfox source and compile it with the CLI. The source owns synthesis, reusable voices, event timing, bounded variation, fixed loudness and playback intent. Output is mono PCM16 WAV at 48 kHz.

Author a voice and phrase

Start with the complete sound syntax example. A voice contains a noise, FM, vocal, chirp or resonator source plus gain and filters. A sequence places named steps and repeats them a finite number of times. Every step explicitly supplies timing, pitch, gain and duration variation ranges. Zero ranges and [1, 1] multipliers disable variation.

Use scalar controls for constants and shared curves for pitch, gain, cutoff, FM index and chirp brightness. Curve time runs from 0 to 1 within each event. Smooth interpolation eases into every knot; log-domain interpolation is useful for frequency. One-shot gain curves must start and end at zero; no hidden fade or automatic normalization corrects the result.

ashfox-model 1
sound bell_pattern {
  duration = 2.4;
  sampleRate = 48000;
  seed = 7349;
  voices = [{
    id = "bell";
    source = {
      kind = "resonator";
      excitation = { kind = "impulse"; };
      modes = [
        { id = "body"; hz = 640; decay = 0.35; gain = 1; },
        { id = "edge"; hz = 1730; decay = 0.18; gain = 0.35; }
      ];
    };
    gain = {
      domain = "linear";
      interpolation = "smooth";
      points = [
        { at = 0; value = 0; },
        { at = 0.01; value = 0.8; },
        { at = 0.7; value = 0.5; },
        { at = 1; value = 0; }
      ];
    };
    highpass = 80;
    lowpass = {
      domain = "log";
      interpolation = "smooth";
      points = [{ at = 0; value = 8000; }, { at = 1; value = 3000; }];
    };
  }];
  sequences = [{
    id = "phrase";
    start = 0.1;
    repeat = { count = 4; period = 0.5; };
    steps = [{
      id = "strike";
      voice = "bell";
      at = 0;
      duration = 0.4;
      gain = 1;
      pitchCents = 0;
      vary = {
        timing = [-0.005, 0.005];
        pitchCents = [-20, 20];
        gain = [0.9, 1.1];
        duration = [0.95, 1.05];
      };
    }];
  }];
  variants = [{ id = "base"; seed = 42; }, { id = "alternate"; seed = 43; }];
  playback = { kind = "loop"; start = 0; end = 2.1; crossfade = 0.08; };
  output = { gainDb = -6; peakDb = -3; };
}

npx --no-install ashfox inspect bell_pattern.ashfox
npx --no-install ashfox build bell_pattern.ashfox --json
npx --no-install ashfox export bell_pattern.ashfox --variant base --output bell.wav

Choose levels and variation

output.gainDb is a fixed multiplier for every variant. peakDb is a rejection ceiling, not a limiter. Lower gain when the compiler reports excess peak; leave headroom for PCM16 rounding. Changing the seed does not guarantee different bytes for a deterministic voice with all variation disabled.

Named sequence, step and voice IDs own independent random streams. Reordering records or appending repetitions preserves existing event randomness. Renaming an ID intentionally changes it. Variation bounds must fit the full timeline, frequency ranges and resource limits, including outcomes not sampled this time.

Make a loop

Set playback = { kind = "loop"; start = 0.4; end = 4.4; crossfade = 0.08; }; in a source whose raw duration covers that interval. The compiler crops after warm-up and bakes the overlap once. A four-second interval with 0.08-second crossfade produces 3.92 seconds, not four. The playback origin shifts by the overlap; an intro or outro should be another asset.

WAV contains only PCM, with no embedded loop instruction. Standalone consumers must enable full-buffer looping. Build receipts and game_assets manifests carry { kind: "loop", startFrame: 0, endFrame: frames }; the game consumer must honor it. One-shots carry { kind: "oneshot" }.

Loop OGG and Minecraft delivery are rejected because those paths cannot preserve the exact loop contract. One-shot OGG still requires FFmpeg with libvorbis. The audio study plays loop WAV directly and offers no OGG download for it.

Listen and inspect

Compare variants at the intended output volume, then compare timbre with matched loudness. Hear an entire phrase and at least 20 loop cycles. Listen for clicks, unnatural repetition, harsh upper harmonics and a dip during crossfade. Check waveform, seam delta, adjacent slopes and spectrogram alongside listening; measurements cannot establish animal or material realism.

The repository's examples/sounds/src contains bird phrases, footsteps, charging spell, modal metal/wood/glass studies and wind/fire/mechanical loops. These are procedural studies requiring audition, not recordings or realism guarantees. bell_pattern.ashfox exercises all five features together.

Older layers, attack, release, contour, sweepSeconds, pitch pairs and output.rmsDb are rejected. Migrate source explicitly; the header remains ashfox-model 1. Use the installed CLI's sound contract fingerprint to check that the executable and source reference agree.