Files
Diabolo/DoOtboy/tools/prepare_soundscape.py

62 lines
2.9 KiB
Python

"""Prepare quiet mono phone-call ambience and foley from local generated takes."""
from pathlib import Path
import hashlib
import json
import sys
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT / "builds/python_audio"))
import numpy as np
import soundfile as sf
work = ROOT / "builds/soundscape"
target = ROOT / "do-otboy/assets/audio/call"
target.mkdir(parents=True, exist_ok=True)
manifest = json.loads((work / "assembly.json").read_text(encoding="utf-8"))
def save(entry, samples, rate):
samples = samples - samples.mean()
rms = np.sqrt(np.mean(samples ** 2))
assert rms > 1e-5 and np.isfinite(samples).all(), entry["id"]
level = -23 if entry["loop"] else -20
samples *= min(10 ** (level / 20) / rms, .45 / np.max(np.abs(samples)))
if entry["loop"]:
width = int(rate * .2)
blend = np.linspace(0, 1, width)
samples[:width] = samples[-width:] * (1 - blend) + samples[:width] * blend
samples = samples[:-width]
else:
width = min(int(rate * .008), len(samples) // 4)
samples[:width] *= np.linspace(0, 1, width)
samples[-width:] *= np.linspace(1, 0, width)
path = target / (entry["id"] + ".wav")
sf.write(path, samples, rate, subtype="PCM_16")
entry.update(file="assets/audio/call/" + path.name, duration=round(len(samples) / rate, 4),
rms_db=round(float(20*np.log10(np.sqrt(np.mean(samples**2)))), 3),
peak_db=round(float(20*np.log10(np.max(np.abs(samples)))), 3),
sha256=hashlib.sha256(path.read_bytes()).hexdigest())
for entry in manifest:
data, rate = sf.read(work / (entry["id"] + ".mp3"), always_2d=True)
save(entry, data.mean(axis=1), rate)
rate = 22050
rng = np.random.default_rng(47)
t = np.arange(rate * 12) / rate
noise = rng.normal(0, 1, len(t))
noise = np.convolve(noise, np.ones(8)/8, mode="same")
line = dict(id="line", loop=True, provider="Original procedural synthesis", description="Very quiet telephone noise floor")
save(line, noise * (.9 + .1*np.sin(2*np.pi*t/12)), rate)
manifest.append(line)
t = np.arange(int(rate * .18)) / rate
click = rng.normal(0, .35, len(t)) * np.exp(-t * 95) + np.sin(2*np.pi*270*t) * np.exp(-t*65)
switch = dict(id="switch", loop=False, provider="Original procedural synthesis", description="Soft power or flashlight switch")
save(switch, click, rate)
manifest.append(switch)
(ROOT / "do-otboy/data/sound_manifest.json").write_text(json.dumps(manifest, ensure_ascii=False, indent=2)+"\n", encoding="utf-8")
print("Prepared", len(manifest), "mono PCM16 sound layers; peaks <= -6.9 dBFS")
voice_levels = {}
for path in (ROOT / "do-otboy/assets/voice").glob("*.wav"):
samples, _ = sf.read(path)
voice_levels[path.stem] = round(float(20*np.log10(max(1e-8, np.sqrt(np.mean(samples**2))))), 3)
(ROOT / "do-otboy/data/voice_levels.json").write_text(json.dumps(voice_levels, indent=2)+"\n", encoding="utf-8")