Upload 5 files
Browse files
app.py
CHANGED
|
@@ -1114,6 +1114,29 @@ def _tts_clip_paths(translated_texts, voice_id, rate, tmp_dir, engine='ms',
|
|
| 1114 |
return tts_paths
|
| 1115 |
|
| 1116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1117 |
def render_retimed_dub(vpath, segments, translated_texts, voice_id, rate, tmp_dir,
|
| 1118 |
total_dur, out_file, engine='ms', speed=0, emotion='',
|
| 1119 |
prog_fn=None):
|
|
@@ -1124,16 +1147,23 @@ def render_retimed_dub(vpath, segments, translated_texts, voice_id, rate, tmp_di
|
|
| 1124 |
- Gaps between segments are kept at 1x.
|
| 1125 |
- Output video length grows accordingly; audio is placed on the new timeline.
|
| 1126 |
"""
|
| 1127 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1128 |
engine=engine, speed=speed, emotion=emotion,
|
| 1129 |
prog_fn=prog_fn)
|
|
|
|
|
|
|
|
|
|
| 1130 |
|
| 1131 |
# 1) Clean clip list (sequential, no overlaps)
|
| 1132 |
clips = [] # (src_start, src_end, clip_path, clip_dur)
|
| 1133 |
prev_end = 0.0
|
| 1134 |
-
for i,
|
| 1135 |
-
s = max(
|
| 1136 |
-
e = min(
|
| 1137 |
if e <= s:
|
| 1138 |
continue
|
| 1139 |
clip = tts_paths[i] if i < len(tts_paths) else None
|
|
|
|
| 1114 |
return tts_paths
|
| 1115 |
|
| 1116 |
|
| 1117 |
+
def _merge_segments(segments, texts):
|
| 1118 |
+
"""
|
| 1119 |
+
Gemini timestamps sometimes overlap. Instead of dropping the overlapped
|
| 1120 |
+
segment's speech, merge it into the previous segment (extend its end and
|
| 1121 |
+
join the text). Returns list of [start, end, text].
|
| 1122 |
+
"""
|
| 1123 |
+
out = []
|
| 1124 |
+
for seg, txt in zip(segments, texts):
|
| 1125 |
+
try:
|
| 1126 |
+
s = float(seg['start']); e = float(seg['end'])
|
| 1127 |
+
except (KeyError, TypeError, ValueError):
|
| 1128 |
+
continue
|
| 1129 |
+
t = (txt or seg.get('text', '') or '').strip()
|
| 1130 |
+
if e <= s or not t:
|
| 1131 |
+
continue
|
| 1132 |
+
if out and s < out[-1][1]:
|
| 1133 |
+
out[-1][1] = max(out[-1][1], e)
|
| 1134 |
+
out[-1][2] = (out[-1][2] + ' ' + t).strip()
|
| 1135 |
+
else:
|
| 1136 |
+
out.append([s, e, t])
|
| 1137 |
+
return out
|
| 1138 |
+
|
| 1139 |
+
|
| 1140 |
def render_retimed_dub(vpath, segments, translated_texts, voice_id, rate, tmp_dir,
|
| 1141 |
total_dur, out_file, engine='ms', speed=0, emotion='',
|
| 1142 |
prog_fn=None):
|
|
|
|
| 1147 |
- Gaps between segments are kept at 1x.
|
| 1148 |
- Output video length grows accordingly; audio is placed on the new timeline.
|
| 1149 |
"""
|
| 1150 |
+
merged = _merge_segments(segments, translated_texts)
|
| 1151 |
+
merged_texts = [m[2] for m in merged]
|
| 1152 |
+
print(f'[dub_retime] segments {len(segments)} -> merged {len(merged)}')
|
| 1153 |
+
|
| 1154 |
+
tts_paths = _tts_clip_paths(merged_texts, voice_id, rate, tmp_dir,
|
| 1155 |
engine=engine, speed=speed, emotion=emotion,
|
| 1156 |
prog_fn=prog_fn)
|
| 1157 |
+
missing = sum(1 for p_ in tts_paths if not p_)
|
| 1158 |
+
if missing:
|
| 1159 |
+
print(f'[dub_retime] TTS missing for {missing}/{len(merged)} segments')
|
| 1160 |
|
| 1161 |
# 1) Clean clip list (sequential, no overlaps)
|
| 1162 |
clips = [] # (src_start, src_end, clip_path, clip_dur)
|
| 1163 |
prev_end = 0.0
|
| 1164 |
+
for i, (m_s, m_e, _m_t) in enumerate(merged):
|
| 1165 |
+
s = max(m_s, prev_end)
|
| 1166 |
+
e = min(m_e, total_dur)
|
| 1167 |
if e <= s:
|
| 1168 |
continue
|
| 1169 |
clip = tts_paths[i] if i < len(tts_paths) else None
|