SoundGame/tools/gen_simon_syn.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""《音序回响》程序合成音效（simon polish round2，research/simon-polish/round1 取证 S2）。

新增 1 件（落 web/simon/audio/sfx/）：
  input_open   「答题窗口开放」标记音：sfx/input_open。confirm 震只标记"播放完毕"，
               占用预告/your_turn 语音链之后才真开放（_toInput）；此前玩家作答被静默
               吞（live-A 行 108-115）。该音为游戏层资产、不走向导层——guide=off /
               iOS（无震动，00 §2.5）下仍有听觉边界。与 6 个手势音色（弹拨/琶音/底鼓/
               木琴双响）及向导 turn earcon（1318.5Hz 单音）错开：双音上行柔铃。

one-shot QC（同 gen_space_syn.py 标准）：去直流 + 首尾 8ms cos² + 真峰值 ≤-3dBFS
（编码后 ebur128 实测，超限自动降档重编）+ 编码后解码回读频谱。

用法：.venv-tts/bin/python tools/gen_simon_syn.py
"""
import os
import subprocess
import sys

import numpy as np
import soundfile as sf

ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
FFMPEG = os.path.join(ROOT, "tools", "bin", "ffmpeg")
OUT = os.path.join(ROOT, "tools", "gen_simon_syn_wav")
DST = os.path.join(ROOT, "web", "simon", "audio", "sfx")
os.makedirs(OUT, exist_ok=True)
os.makedirs(DST, exist_ok=True)

sr = 44100
ONE_SHOT_TP_CEIL = -3.0
SAMPLE_PEAK = 0.66
FADE_N = int(sr * 0.008)


def dc(y):
    return y - np.mean(y)


def fade(y):
    w = np.sin(np.linspace(0, np.pi / 2, FADE_N)) ** 2
    y = y.copy()
    y[:FADE_N] *= w
    y[-FADE_N:] *= w[::-1]
    return y


def norm(y, peak=SAMPLE_PEAK):
    m = np.max(np.abs(y))
    return (y / m * peak) if m > 0 else y


def stereo(y):
    return np.stack([y, y], axis=1).astype(np.float32)


def bell(freq, dur, gain=1.0, shimmer=0.18):
    """柔铃单音：基频 + 弱二次泛音，快起双段缓释（同向导 earcon 包络思路）。"""
    n = int(dur * sr)
    t = np.arange(n) / sr
    y = np.sin(2 * np.pi * freq * t) + shimmer * np.sin(2 * np.pi * freq * 2 * t)
    env = np.minimum(1.0, t / 0.008)
    env *= np.where(t < dur * 0.4,
                    np.exp(-t * 3.0),
                    np.exp(-dur * 0.4 * 3.0) * np.exp(-(t - dur * 0.4) * 9.0))
    return y * env * gain


def input_open():
    """双音上行柔铃 B5→D6（0.35s）：「开放作答」。不与 tone_doubletap（同音双响）
    或 turn earcon（1318.5Hz 单音）同质。"""
    y = np.zeros(int(sr * 0.38))
    for at, f in [(0.0, 987.77), (0.14, 1174.66)]:
        seg = bell(f, 0.3, gain=0.9)
        n = min(len(seg), len(y) - int(at * sr))
        y[int(at * sr):int(at * sr) + n] += seg[:n]
    return dc(y)


ONE_SHOTS = {
    "input_open": input_open,
}


# ---------------------------------------------------------------- 编码 + 真峰值校验
def encode(wav_path, base):
    for args, ext in [(["-c:a", "libvorbis", "-q:a", "3"], ".ogg"),
                      (["-c:a", "aac", "-b:a", "128k"], ".m4a")]:
        subprocess.run([FFMPEG, "-y", "-v", "error", "-i", wav_path,
                        "-ar", "44100", *args, base + ext], check=True)


def true_peak_db(path):
    r = subprocess.run([FFMPEG, "-i", path, "-af", "ebur128=peak=true", "-f", "null", "-"],
                       capture_output=True, text=True)
    peaks = [float(line.split("Peak:")[1].strip().split()[0])
             for line in r.stderr.splitlines() if "Peak:" in line and "dBFS" in line]
    return max(peaks) if peaks else float("nan")


def check_tp(base):
    return max(true_peak_db(base + ".ogg"), true_peak_db(base + ".m4a"))


def spectral_report(base):
    """编码后解码回读：≥3kHz/≥5kHz 能量占比（同 space polish 复测口径）。"""
    r = subprocess.run([FFMPEG, "-v", "error", "-i", base + ".ogg", "-f", "f32le", "-ac", "1", "-ar", "44100", "-"],
                       capture_output=True)
    y = np.frombuffer(r.stdout, dtype=np.float32)
    Y = np.abs(np.fft.rfft(y * np.hanning(len(y))))
    fr = np.fft.rfftfreq(len(y), 1 / sr)
    e = Y ** 2
    tot = e.sum() + 1e-12
    return e[fr >= 3000].sum() / tot * 100, e[fr >= 5000].sum() / tot * 100, fr[np.argmax(Y)] / 1000


def main():
    report = []
    problems = []
    for fid, fn in ONE_SHOTS.items():
        wav_path = os.path.join(OUT, fid + ".wav")
        base = os.path.join(DST, fid)
        y = fade(norm(dc(fn())))
        sf.write(wav_path, stereo(y), sr)
        encode(wav_path, base)
        tp = check_tp(base)
        if tp > ONE_SHOT_TP_CEIL:
            gain = 10 ** ((ONE_SHOT_TP_CEIL - tp) / 20) * 0.98
            sf.write(wav_path, stereo(y * gain), sr)
            encode(wav_path, base)
            tp2 = check_tp(base)
            report.append((fid, len(y) / sr, tp, f"降 {20 * np.log10(gain):.1f}dB 重编 -> {tp2:.2f}"))
            tp = tp2
        else:
            report.append((fid, len(y) / sr, tp, ""))
        if tp > ONE_SHOT_TP_CEIL:
            problems.append(f"{fid}: 真峰值 {tp:.2f} dBFS 仍超 {ONE_SHOT_TP_CEIL}")
    print(f"{'id':14s} {'dur(s)':>7s} {'TP(dBFS)':>9s} {'%>3k':>7s} {'%>5k':>7s} {'peakkHz':>8s}  note")
    for fid, dur, tp, note in report:
        e3, e5, pk = spectral_report(os.path.join(DST, fid))
        print(f"{fid:14s} {dur:7.2f} {tp:9.2f} {e3:7.1f} {e5:7.1f} {pk:8.2f}  {note}")
        if e5 > 15.0:
            problems.append(f"{fid}: 5kHz+ 能量 {e5:.1f}% 仍偏高")
    if problems:
        print("PROBLEMS:")
        for p in problems:
            print(" ", p)
        return 1
    print(f"ALL OK: {len(ONE_SHOTS)} one-shot -> {DST}")
    return 0


if __name__ == "__main__":
    sys.exit(main())