Open-source AI podcast clipper. Generate vertical clips with face tracking and burned-in captions. CLI, MCP server, web app.
Python ai-toolsauto-captionscaptionscliffmpegmcpmcp-serveropusclip-alternativepodcast-clipspodcastsshortform-videovertical-videowhisper
Window YAMNet inference so long audio can't block the worker
Nika Siradze0c82f0ffa49c
diff --git a/backend/services/audio_events.py b/backend/services/audio_events.pyindex ce1a265..c6b862c 100644--- a/backend/services/audio_events.py+++ b/backend/services/audio_events.py@@ -119,13 +119,27 @@ def extract_audio_events(video_path: str) -> list[dict]:return []input_name = session.get_inputs()[0].name- scores = session.run(None, {input_name: waveform})[0] # [frames, 521]- n_frames = scores.shape[0]- if n_frames == 0:- return []- duration = waveform.size / 16000.0- hop = duration / n_frames+ # Windowed so one multi-hour session.run() can't block the worker unboundedly.+ window = 16000 * 300+ parts: list[np.ndarray] = []+ times: list[float] = []+ for offset in range(0, waveform.size, window):+ chunk = waveform[offset:offset + window]+ if chunk.size < 16000:+ continue+ chunk_scores = session.run(None, {input_name: chunk})[0] # [frames, 521]+ if chunk_scores.shape[0] == 0:+ continue+ hop = (chunk.size / 16000.0) / chunk_scores.shape[0]+ base = offset / 16000.0+ times.extend(base + f * hop for f in range(chunk_scores.shape[0]))+ parts.append(chunk_scores)++ if not parts:+ return []+ scores = np.concatenate(parts, axis=0)+ n_frames = scores.shape[0]def channel_max(keys: list[int]) -> np.ndarray:return scores[:, keys].max(axis=1) if keys else np.zeros(n_frames)@@ -137,7 +151,7 @@ def extract_audio_events(video_path: str) -> list[dict]:return [{- "time": round(i * hop, 2),+ "time": round(times[i], 2),"laughter": round(float(laughter[i]), 3),"cheering": round(float(cheering[i]), 3),"screaming": round(float(screaming[i]), 3),