# -*- coding: utf-8 -*- """ Text 1 讲解视频 —— 配音与时间轴生成 流程: 1. 解析 讲解脚本_Text1.md(## 场景key | 场景名 + 以 "- " 开头的口播行) 2. 用 edge-tts 逐场景合成配音(走新端点 api.msedgeservices.com) 3. 用返回的 SentenceBoundary 时间戳,按字符对齐算出每一条字幕的精确起止 4. 输出 public/vo/.mp3 与 src/timing.generated.ts """ import asyncio import os import re import subprocess import sys import time as _time # ---------- edge-tts 新端点补丁(2026-10 起旧域名返回 403)---------- import edge_tts import edge_tts.constants as K K.WSS_URL = ( "wss://api.msedgeservices.com/tts/cognitiveservices/websocket/v1" f"?Ocp-Apim-Subscription-Key={K.TRUSTED_CLIENT_TOKEN}" ) import edge_tts.communicate as CM CM.WSS_URL = K.WSS_URL HERE = os.path.dirname(os.path.abspath(__file__)) PROJ = os.path.join(HERE, "remotion-text1") SCRIPT_MD = os.path.join(HERE, "讲解脚本_Text1.md") VO_DIR = os.path.join(PROJ, "public", "vo") OUT_TS = os.path.join(PROJ, "src", "timing.generated.ts") FFMPEG = os.path.join(HERE, ".venv", "Lib", "site-packages", "imageio_ffmpeg", "binaries", "ffmpeg-win-x86_64-v7.1.exe") VOICE = "zh-CN-YunjianNeural" RATE = "-4%" FPS = 30 LEAD_PAD = 0 # 场景开头留白(帧) TAIL_PAD = 0.7 # 场景结尾留白(秒),给转场淡出 # ---------------- 1. 解析脚本 ---------------- def parse_script(path): scenes = [] cur = None with open(path, encoding="utf-8") as f: for raw in f: line = raw.rstrip("\n") if line.startswith("## "): head = line[3:].strip() if "|" in head: key, label = [x.strip() for x in head.split("|", 1)] else: key, label = head, head cur = {"key": key, "label": label, "lines": []} scenes.append(cur) elif line.startswith("- ") and cur is not None: t = line[2:].strip() if t: cur["lines"].append(t) return [s for s in scenes if s["lines"]] # ---------------- 2. 时间戳对齐 ---------------- NORM_RE = re.compile(r"[^0-9A-Za-z\u4e00-\u9fff]+") def norm(s: str) -> str: """只保留字母、数字与汉字,去掉空白与标点。""" return NORM_RE.sub("", s).lower() def build_char_map(full_text: str): """返回 (full_norm, 映射:原文字符下标 -> 归一化下标)。""" out = [] idx_of = [] for ch in full_text: n = norm(ch) for _ in n: idx_of.append(len(out)) if n: out.append(n) else: idx_of.append(len(out)) # 标点/空白:映射到当前位置 idx_of.append(len(out)) return "".join(out), idx_of def cue_times(full_text, cue_spans, boundaries, audio_dur): """把每条字幕的字符区间换算成时间(秒)。 cue_spans: [(start_char, end_char)] 相对 full_text boundaries: [(norm_text, start_sec, end_sec)] """ full_norm, idx_of = build_char_map(full_text) n = len(full_norm) times = [None] * (n + 1) p = 0 matched = 0 for btxt, bstart, bend in boundaries: bn = norm(btxt) if not bn: continue pos = full_norm.find(bn, max(0, p - 2)) if pos < 0: pos = full_norm.find(bn) if pos < 0: continue matched += len(bn) L = len(bn) for k in range(L): t = bstart + (bend - bstart) * (k / L if L else 0) gi = pos + k if gi <= n and (times[gi] is None): times[gi] = t if pos + L <= n: times[pos + L] = bend if times[pos + L] is None else times[pos + L] p = pos + L coverage = matched / n if n else 0 # 已知锚点线性插值补全 known = [i for i, t in enumerate(times) if t is not None] if known: if times[0] is None: times[0] = 0.0 for a, b in zip(known, known[1:]): if b - a > 1: t0, t1 = times[a], times[b] for k in range(a + 1, b): times[k] = t0 + (t1 - t0) * (k - a) / (b - a) tail = times[known[-1]] for i in range(known[-1] + 1, n + 1): times[i] = tail out = [] for (cs, ce) in cue_spans: ns = idx_of[max(0, min(cs, len(idx_of) - 1))] ne = idx_of[max(0, min(ce, len(idx_of) - 1))] t0 = times[min(ns, n)] if ns <= n else 0.0 t1 = times[min(ne, n)] if ne <= n else audio_dur if t0 is None: t0 = 0.0 if t1 is None or t1 <= t0: t1 = min(audio_dur, t0 + 0.4) out.append((max(0.0, t0), max(t0 + 0.2, t1))) return out, coverage # ---------------- 3. 音频工具 ---------------- def ffprobe_duration(path): r = subprocess.run([FFMPEG, "-hide_banner", "-i", path], capture_output=True, text=True, errors="replace") m = re.search(r"Duration: (\d+):(\d+):(\d+\.\d+)", r.stderr) if not m: raise RuntimeError("无法读取时长: " + path + "\n" + r.stderr[-500:]) h, mi, s = m.group(1), m.group(2), m.group(3) return int(h) * 3600 + int(mi) * 60 + float(s) # ---------------- 4. 合成 ---------------- async def synth(text, out_path, tries=4): last = None for i in range(tries): try: c = edge_tts.Communicate(text, VOICE, rate=RATE) buf = b"" bounds = [] async for ch in c.stream(): t = ch.get("type", "") if t == "audio": buf += ch["data"] elif t.endswith("Boundary"): bounds.append((ch.get("text", ""), ch.get("offset", 0) / 1e7, (ch.get("offset", 0) + ch.get("duration", 0)) / 1e7)) if len(buf) < 2000: raise RuntimeError(f"音频过短 ({len(buf)} bytes)") if not bounds: raise RuntimeError("未收到时间戳事件") with open(out_path, "wb") as f: f.write(buf) return bounds except Exception as e: # noqa: BLE001 last = e print(f" 重试 {i+1}/{tries}: {type(e).__name__} {str(e)[:80]}") await asyncio.sleep(2.0 + i * 1.5) raise RuntimeError(f"合成失败: {last}") def ts_escape(s): return s.replace("\\", "\\\\").replace("'", "\\'") async def main(): scenes = parse_script(SCRIPT_MD) os.makedirs(VO_DIR, exist_ok=True) print(f"共 {len(scenes)} 个场景,语音={VOICE},语速={RATE}\n") result = [] cursor_frames = 0 for idx, sc in enumerate(scenes, 1): key, label, lines = sc["key"], sc["label"], sc["lines"] full = "".join(lines) # 每条字幕的字符区间 spans, p = [], 0 for ln in lines: spans.append((p, p + len(ln))) p += len(ln) out_mp3 = os.path.join(VO_DIR, f"{key}.mp3") print(f"[{idx:02d}/{len(scenes)}] {key:<11} {len(lines):>2} 句 / {len(full):>4} 字 ...") t0 = _time.time() bounds = await synth(full, out_mp3) dur = ffprobe_duration(out_mp3) cts, cov = cue_times(full, spans, bounds, dur) print(f" 音频 {dur:.2f}s · 边界 {len(bounds)} · 对齐覆盖 {cov*100:.0f}% · 耗时 {_time.time()-t0:.1f}s") start_frame = cursor_frames scene_dur = round((dur + TAIL_PAD) * FPS) cues = [] for (a, b), txt in zip(cts, lines): cues.append({ "text": txt, "start": start_frame + round(a * FPS), "end": start_frame + round(min(b, dur) * FPS), }) result.append({ "key": key, "label": label, "audio": f"vo/{key}.mp3", "start": start_frame, "dur": scene_dur, "cues": cues, }) cursor_frames = start_frame + scene_dur # ---- 写 TS ---- L = [] L.append("// 本文件由 build_voice.py 自动生成,请勿手工修改。") L.append("export type Cue = { text: string; start: number; end: number };") L.append("export type SceneTiming = {") L.append(" key: string; label: string; audio: string;") L.append(" start: number; dur: number; cues: Cue[];") L.append("};") L.append("") L.append("export const SCENES: SceneTiming[] = [") for r in result: L.append(" {") L.append(f" key: '{ts_escape(r['key'])}',") L.append(f" label: '{ts_escape(r['label'])}',") L.append(f" audio: '{r['audio']}',") L.append(f" start: {r['start']},") L.append(f" dur: {r['dur']},") L.append(" cues: [") for c in r["cues"]: L.append(" { text: '%s', start: %d, end: %d }," % (ts_escape(c["text"]), c["start"], c["end"])) L.append(" ],") L.append(" },") L.append("];") L.append("") L.append(f"export const TOTAL_FRAMES = {cursor_frames};") L.append(f"export const TOTAL_SECONDS = {cursor_frames / FPS:.2f};") L.append("") with open(OUT_TS, "w", encoding="utf-8") as f: f.write("\n".join(L)) print(f"\n完成:{len(result)} 个场景,总时长 {cursor_frames/FPS/60:.1f} 分钟({cursor_frames} 帧)") print("输出:", OUT_TS) for r in result: print(f" {r['key']:<11} start={r['start']:>6} dur={r['dur']:>6} ({r['dur']/FPS:>5.1f}s) cues={len(r['cues'])}") if __name__ == "__main__": sys.exit(asyncio.run(main()))