Something went wrong. Try again.
Monorepo for Aesthetic.Computer aesthetic.computer
Something went wrong. Try again.
Python
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122#!/usr/bin/env python3"""worldsing.py — single-note WORLD-vocoder pitch-correction tuned for a"sung instrument" tone: aggressive aperiodicity scaling (less breath),unvoiced attenuation (no consonant air between vowels), and an optionalspectral tilt to push toward a hollow tube/vocoder character.
Local to pop/marimba/. The shared pop/bin/pitchsnap_world.py is builtfor big-pictures vocals where natural breath/consonants matter; thishelper is for instrument-blend baas/woos where they don't.
Pipeline: audio → harvest/stonemask (f0) → cheaptrick (spectral envelope = formants) → d4c (aperiodicity = breath/noise) → SCALE ap down (less air) → REPLACE f0 with constant target pitch → synthesize → ATTENUATE unvoiced regions (drop in/out breath)
Usage: worldsing.py in.wav out.wav --note A3 \\ --ap-scale 0.10 # 1.0 = WORLD default airy, 0.0 = no breath --unvoiced-gain 0.05 # 1.0 = preserve consonants, 0.0 = mute them"""import argparse, sysimport numpy as npimport soundfile as sfimport pyworld as pw
NOTE_SEMI = {"c": 0, "d": 2, "e": 4, "f": 5, "g": 7, "a": 9, "b": 11}
def note_to_midi(s): s = s.strip().lower() semi = NOTE_SEMI[s[0]] i = 1 if i < len(s) and s[i] in "#b": semi += 1 if s[i] == "#" else -1 i += 1 return 12 * (int(s[i:]) + 1) + semi
def midi_to_hz(m): return 440.0 * (2.0 ** ((m - 69.0) / 12.0))
def main(): p = argparse.ArgumentParser() p.add_argument("in_wav") p.add_argument("out_wav") p.add_argument("--note", required=True, help="Target note name, e.g. A3, Eb4, F#5") p.add_argument("--ap-scale", type=float, default=0.10, help="Multiply aperiodicity by this. 1.0 = full breath " "(default WORLD), 0.0 = pure tonal. 0.10 = hollow.") p.add_argument("--unvoiced-gain", type=float, default=0.05, help="Gain applied to unvoiced/breath regions. " "1.0 = preserve, 0.0 = mute completely.") args = p.parse_args()
x, fs = sf.read(args.in_wav, dtype="float64") if x.ndim > 1: x = x.mean(axis=1)
f0_floor = 90.0 f0_raw, t = pw.harvest(x, fs, f0_floor=f0_floor, f0_ceil=600.0, frame_period=5.0) f0 = pw.stonemask(x, f0_raw, t, fs) fft_size = pw.get_cheaptrick_fft_size(fs, f0_floor=f0_floor) sp = pw.cheaptrick(x, f0, t, fs, fft_size=fft_size, f0_floor=f0_floor) ap = pw.d4c(x, f0, t, fs, fft_size=fft_size)
# Aperiodicity scale — the main "less air" knob. ap is in [0, 1] # per (freq_bin, frame); scaling it down makes WORLD synthesise a # more sinusoidal / tonal output instead of breathy speech. ap = np.clip(ap * args.ap_scale, 1e-6, 1.0)
# Replace f0 with constant target across all frames (the source is # always a single short syllable, no need for a curve). target_hz = midi_to_hz(note_to_midi(args.note)) n_frames = len(t) voiced = f0 > 0 f0_synth = np.full(n_frames, target_hz, dtype=np.float64)
y = pw.synthesize(f0_synth, sp, ap, fs, frame_period=5.0)
# Build voiced/unvoiced amplitude mask in time domain with 5ms ramps # at boundaries so consonant edges crossfade smoothly. samples_per_frame = int(round(fs * 5.0 / 1000.0)) voiced_mask = np.repeat(voiced.astype(np.float64), samples_per_frame) if len(voiced_mask) < len(y): voiced_mask = np.pad(voiced_mask, (0, len(y) - len(voiced_mask)), mode="edge") voiced_mask = voiced_mask[:len(y)] ramp = int(0.005 * fs) if ramp > 1: edges = np.diff(voiced_mask.astype(np.int8)) for idx in np.where(edges == 1)[0]: for k in range(ramp): pos = idx + 1 + k if pos < len(voiced_mask): voiced_mask[pos] *= 0.5 - 0.5 * np.cos(np.pi * (k + 1) / ramp) for idx in np.where(edges == -1)[0]: for k in range(ramp): pos = idx - k if pos >= 0: voiced_mask[pos] *= 0.5 - 0.5 * np.cos(np.pi * (k + 1) / ramp)
# Composite: WORLD-synth at full level in voiced regions; original # at `unvoiced-gain` in unvoiced. unvoiced-gain=0.0 leaves only the # sung vowels, no breath/consonants. n = min(len(y), len(x), len(voiced_mask)) y_out = (voiced_mask[:n] * y[:n] + (1.0 - voiced_mask[:n]) * x[:n] * args.unvoiced_gain) sf.write(args.out_wav, y_out.astype(np.float32), fs)
voiced_pct = 100.0 * np.mean(voiced) print(f" worldsing · {args.note} ({target_hz:.1f}Hz) · " f"ap×{args.ap_scale} · unv×{args.unvoiced_gain} · " f"{voiced_pct:.0f}% voiced · {n} samples") return 0
if __name__ == "__main__": sys.exit(main())