apps/api/app/tasks/lipsync_tts.py

@@ -334,6 +334,7 @@ def tts_synthesize_and_submit(
         db.commit()
 
         # 2.5 计算精确句子时间戳（基于 TTS 音频静音检测）
+        # 直接复用步骤 2 已下载到内存的 audio_data，避免重新从 OSS 下载（私有桶未签名会失败）
         import os as _os
 
         _st_tmp_path = None
@@ -341,49 +342,63 @@ def tts_synthesize_and_submit(
             import subprocess as _sp
             import tempfile as _tmpf
 
-            # 下载音频用于探测时长和静音检测
-            if isinstance(job.audio_url, str) and job.audio_url:
-                from packages.shared.url_security import safe_download_bytes as _sdl
-
-                _audio_bytes = _sdl(job.audio_url, purpose="sentence_timings", timeout=30.0)
+            if not audio_data:
+                logger.warning("[lipsync_tts] 无音频数据，跳过句子时间戳计算: job_id=%s", job_id)
             else:
-                _audio_bytes = audio_data
+                # 写入临时文件供 ffprobe/ffmpeg 使用
+                with _tmpf.NamedTemporaryFile(suffix=".mp3", delete=False) as _atmp:
+                    _atmp.write(audio_data)
+                    _st_tmp_path = _atmp.name
 
-            # ffprobe 获取音频时长
-            with _tmpf.NamedTemporaryFile(suffix=".mp3", delete=False) as _atmp:
-                _atmp.write(_audio_bytes)
-                _st_tmp_path = _atmp.name
+                # ffprobe 获取音频时长
+                _probe_result = _sp.run(
+                    [
+                        "ffprobe",
+                        "-v",
+                        "error",
+                        "-show_entries",
+                        "format=duration",
+                        "-of",
+                        "default=noprint_wrappers=1:nokey=1",
+                        _st_tmp_path,
+                    ],
+                    capture_output=True,
+                    text=True,
+                    timeout=10,
+                )
+                _audio_duration = float(_probe_result.stdout.strip()) if _probe_result.stdout.strip() else 0.0
+                logger.info(
+                    "[lipsync_tts] 音频时长探测: job_id=%s duration=%.2f probe_stdout=%s probe_stderr=%s",
+                    job_id,
+                    _audio_duration,
+                    _probe_result.stdout.strip()[:50],
+                    _probe_result.stderr.strip()[:100] if _probe_result.stderr else "",
+                )
 
-            _probe_result = _sp.run(
-                [
-                    "ffprobe",
-                    "-v",
-                    "error",
-                    "-show_entries",
-                    "format=duration",
-                    "-of",
-                    "default=noprint_wrappers=1:nokey=1",
-                    _st_tmp_path,
-                ],
-                capture_output=True,
-                text=True,
-                timeout=10,
-            )
-            _audio_duration = float(_probe_result.stdout.strip()) if _probe_result.stdout.strip() else 0.0
-
-            if _audio_duration > 0:
-                _timings = _compute_sentence_timings(_audio_bytes, script_text, _audio_duration)
-                if _timings:
-                    job.sentence_timings = _timings
-                    logger.info(
-                        "[lipsync_tts] 句子时间戳已计算: job_id=%s sentences=%d duration=%.1f",
+                if _audio_duration > 0:
+                    _timings = _compute_sentence_timings(audio_data, script_text, _audio_duration)
+                    if _timings:
+                        job.sentence_timings = _timings
+                        logger.info(
+                            "[lipsync_tts] 句子时间戳已计算: job_id=%s sentences=%d duration=%.1f",
+                            job_id,
+                            len(_timings),
+                            _audio_duration,
+                        )
+                    else:
+                        logger.warning("[lipsync_tts] 句子时间戳计算返回空结果: job_id=%s", job_id)
+                else:
+                    logger.warning(
+                        "[lipsync_tts] ffprobe 未获取到有效时长，跳过句子时间戳: job_id=%s stdout=%s stderr=%s",
                         job_id,
-                        len(_timings),
-                        _audio_duration,
+                        _probe_result.stdout.strip()[:100],
+                        _probe_result.stderr.strip()[:200] if _probe_result.stderr else "",
                     )
-            db.commit()
+                db.commit()
         except Exception as _st_err:
-            logger.warning("[lipsync_tts] 句子时间戳计算失败（不影响主流程）: job_id=%s err=%s", job_id, _st_err)
+            logger.warning(
+                "[lipsync_tts] 句子时间戳计算失败（不影响主流程）: job_id=%s err=%s", job_id, _st_err, exc_info=True
+            )
         finally:
             if _st_tmp_path:
                 try:

apps/web/src/pages/ai-avatar/utils/sentences.ts

@@ -42,9 +42,10 @@ export function splitScriptIntoSentences(
   let accChar = 0
   const totalChars = rawParts.reduce((sum, p) => sum + p.replace(/\s/g, "").length, 0)
   // 如果后端返回了完整的时间戳（至少有一个有效结束时间），使用精确时间
+  // 需要满足：数组长度与分句数一致，且至少有一个结束时间 > 0（防御全0的异常数据）
   const hasBackendTimings =
     sentenceTimings &&
-    sentenceTimings.length > 0 &&
+    sentenceTimings.length >= rawParts.length &&
     sentenceTimings.some((t) => (t.end_time ?? 0) > 0)
 
   // 若有后端时间戳，取音频总时长；否则用外部传入的 outputDuration 做字数比例降级