#!/usr/bin/env python3 """xv 配音:从 30_shots.json 取 line 合成,实测时长回填 duration。 用法: python3 tools/tts.py projects/EPxx/30_shots.json 说明: 不加快语速硬塞;duration = max(标称片长, 实测语音 + 0.6s 尾留白)。 """ import argparse import asyncio import json import os import re import subprocess import sys def _find_root(): """工作空间根:MANJU_ROOT > cwd 向上找 config/pipeline.json > 脚本上级。""" env = os.environ.get("MANJU_ROOT") if env: return os.path.abspath(os.path.expanduser(env)) d = os.getcwd() while True: if os.path.isfile(os.path.join(d, "config", "pipeline.json")): return d nd = os.path.dirname(d) if nd == d: break d = nd return os.path.dirname(os.path.dirname(os.path.abspath(__file__))) ROOT = _find_root() def ffmpeg_bin(): import imageio_ffmpeg return imageio_ffmpeg.get_ffmpeg_exe() def audio_duration(path): r = subprocess.run([ffmpeg_bin(), "-hide_banner", "-i", str(path)], stdout=subprocess.DEVNULL, stderr=subprocess.PIPE, text=True) m = re.search(r"Duration:\s*(\d+):(\d+):([\d.]+)", r.stderr) if not m: return None return int(m.group(1)) * 3600 + int(m.group(2)) * 60 + float(m.group(3)) async def synth(text, out, voice, rate): import edge_tts await edge_tts.Communicate(text, voice, rate=rate).save(str(out)) def main(): ap = argparse.ArgumentParser() ap.add_argument("shots", help="30_shots.json 路径") ap.add_argument("--out-dir", default=None, help="默认 work//audio") ap.add_argument("--voice", default=None) ap.add_argument("--rate", default="+0%") ap.add_argument("--no-writeback", action="store_true") a = ap.parse_args() shots_path = os.path.abspath(a.shots) cfg = json.load(open(os.path.join(ROOT, "config", "pipeline.json"), encoding="utf-8")) voice = a.voice or cfg["tts"]["voice"] nominal = cfg["videoModel"]["nominalSeconds"] data = json.load(open(shots_path, encoding="utf-8")) ep = data["episode"] out_dir = os.path.abspath(a.out_dir) if a.out_dir else os.path.join(ROOT, "work", ep, "audio") os.makedirs(out_dir, exist_ok=True) print("音色: %s | 语速: %s | 输出: %s" % (voice, a.rate, out_dir), flush=True) durations = {} for i, s in enumerate(data["shots"], 1): sid = s["shotId"] out = os.path.join(out_dir, "vo_%02d.mp3" % i) if not (os.path.exists(out) and os.path.getsize(out) > 1000): asyncio.run(synth(s["line"], out, voice, a.rate)) d = audio_duration(out) if d is None: print("ERROR: 无法解析时长 %s" % out, file=sys.stderr) return 1 target = round(max(nominal, d + 0.6), 2) durations[sid] = {"audio": round(d, 2), "target": target, "voice": voice, "file": os.path.relpath(out, ROOT)} print(" %s vo=%5.2fs -> duration=%5.2fs | %s" % (sid, d, target, s["line"]), flush=True) with open(os.path.join(out_dir, "durations.json"), "w", encoding="utf-8") as f: json.dump(durations, f, ensure_ascii=False, indent=2) if not a.no_writeback: for s in data["shots"]: s["duration"] = durations[s["shotId"]]["target"] data["targetDuration"] = round(sum(s["duration"] for s in data["shots"]), 2) with open(shots_path, "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2) print("已回填 30_shots.json: 总时长 %.2fs" % data["targetDuration"], flush=True) return 0 if __name__ == "__main__": sys.exit(main())