#!/usr/bin/env python3 """ Clip Segmenter - Find the best clip-worthy segments from Whisper transcripts Uses Claude API to analyze transcripts and identify standalone clips. Usage: python3 clip_segmenter.py --transcript path/to/transcript.json --output path/to/segments.json python3 clip_segmenter.py --transcript-dir transcripts/ --output-dir segments/ """ import argparse import json import os import sys from pathlib import Path import anthropic import re from datetime import datetime def load_transcript(json_file): """Load Whisper JSON transcript""" with open(json_file, 'r') as f: data = json.load(f) return data def format_timestamp(seconds): """Convert seconds to HH:MM:SS format""" hours = int(seconds // 3600) minutes = int((seconds % 3600) // 60) secs = int(seconds % 60) return f"{hours:02d}:{minutes:02d}:{secs:02d}" def create_full_transcript_text(whisper_data): """Create a readable transcript with timestamps for Claude""" segments = whisper_data.get('segments', []) transcript_lines = [] for segment in segments: start = segment['start'] text = segment['text'].strip() timestamp = format_timestamp(start) transcript_lines.append(f"[{timestamp}] {text}") return '\n'.join(transcript_lines) def analyze_with_claude(transcript_text, episode_title, anthropic_client, model="claude-haiku-4-5-20250514", max_segments=5, min_hook_strength=6): """Send transcript to Claude for segment analysis""" prompt = f"""You are analyzing a podcast transcript to identify the best clip-worthy segments. Episode: {episode_title} TRANSCRIPT: {transcript_text} Please identify {max_segments} standalone segments that would work as viral clips. Each segment should: - Be 3-15 minutes long - Have a clear hook/opening that grabs attention - Contain a complete thought, story, or framework - Feel satisfying as a standalone watch - Have viral potential (contrarian takes, practical advice, compelling stories) For each segment, provide: - start_time: timestamp in seconds (not HH:MM:SS) - end_time: timestamp in seconds - suggested_title: Catchy, clickbait-worthy title (50 chars max) - one_line_description: What the clip is about - hook_strength: 1-10 rating for how compelling the opening is (only include if >= {min_hook_strength}) - key_topics: 2-3 main topics covered Return ONLY a valid JSON array like this: [ {{ "start_time": 65, "end_time": 420, "suggested_title": "Why 99% of Founders Fail at AI", "one_line_description": "A breakdown of the critical mistake most entrepreneurs make when implementing AI", "hook_strength": 9, "key_topics": ["AI implementation", "founder mistakes", "business strategy"] }} ] Focus on segments with strong hooks, practical advice, and contrarian or surprising insights.""" try: response = anthropic_client.messages.create( model=model, max_tokens=4000, messages=[{"role": "user", "content": prompt}] ) response_text = response.content[0].text.strip() # Try to extract JSON from response json_match = re.search(r'\[.*\]', response_text, re.DOTALL) if json_match: json_str = json_match.group(0) segments = json.loads(json_str) return segments else: print(f"Warning: No JSON found in Claude response for {episode_title}") print("Response:", response_text[:200]) return [] except Exception as e: print(f"Error calling Claude API: {e}") return [] def process_single(transcript_file, output_file, episode_title, client, model="claude-haiku-4-5-20250514", max_segments=5, min_hook_strength=6): """Process a single transcript file.""" print(f"\nProcessing: {transcript_file}") whisper_data = load_transcript(transcript_file) if not episode_title: episode_title = Path(transcript_file).stem transcript_text = create_full_transcript_text(whisper_data) print(f"Transcript length: {len(transcript_text)} chars") segments = analyze_with_claude( transcript_text, episode_title, client, model=model, max_segments=max_segments, min_hook_strength=min_hook_strength ) print(f"Found {len(segments)} segments") # Add metadata for segment in segments: segment['episode_file'] = Path(transcript_file).stem segment['video_file'] = Path(transcript_file).stem + '.mp4' # Save segments output_path = Path(output_file) output_path.parent.mkdir(parents=True, exist_ok=True) with open(output_path, 'w') as f: json.dump(segments, f, indent=2) print(f"Saved to: {output_path}") # Print summary for i, seg in enumerate(segments, 1): duration = seg['end_time'] - seg['start_time'] print(f" {i}. {seg['suggested_title']} ({duration:.0f}s, hook: {seg['hook_strength']}/10)") return segments def main(): parser = argparse.ArgumentParser(description="Find clip-worthy segments from Whisper transcripts") parser.add_argument("--transcript", help="Path to a single Whisper JSON transcript") parser.add_argument("--transcript-dir", help="Directory of Whisper JSON transcripts") parser.add_argument("--output", help="Output path for segment JSON (single file mode)") parser.add_argument("--output-dir", help="Output directory for segment JSONs (batch mode)") parser.add_argument("--episode-title", help="Episode title (optional)") parser.add_argument("--model", default="claude-haiku-4-5-20250514", help="Claude model to use") parser.add_argument("--max-segments", type=int, default=5, help="Max clips per episode (default: 5)") parser.add_argument("--min-hook-strength", type=int, default=6, help="Min hook score to include (default: 6)") args = parser.parse_args() if not args.transcript and not args.transcript_dir: parser.error("Provide --transcript or --transcript-dir") # Get Anthropic client api_key = os.environ.get("ANTHROPIC_API_KEY") or os.environ.get("ANTHROPIC_KEY") if not api_key: print("Error: Set ANTHROPIC_API_KEY environment variable") sys.exit(1) client = anthropic.Anthropic(api_key=api_key) if args.transcript: # Single file mode output = args.output or str(Path(args.transcript).parent / f"{Path(args.transcript).stem}_segments.json") process_single( args.transcript, output, args.episode_title, client, model=args.model, max_segments=args.max_segments, min_hook_strength=args.min_hook_strength ) else: # Batch mode transcript_dir = Path(args.transcript_dir) output_dir = Path(args.output_dir or "segments") output_dir.mkdir(parents=True, exist_ok=True) transcript_files = list(transcript_dir.glob("*.json")) print(f"Found {len(transcript_files)} transcripts to process") all_segments = [] for transcript_file in transcript_files: output_file = output_dir / f"{transcript_file.stem}_segments.json" segments = process_single( str(transcript_file), str(output_file), None, client, model=args.model, max_segments=args.max_segments, min_hook_strength=args.min_hook_strength ) all_segments.extend(segments) # Save combined segments combined_file = output_dir / "all_segments.json" with open(combined_file, 'w') as f: json.dump(all_segments, f, indent=2) print(f"\n✅ Segmentation complete!") print(f"Total segments found: {len(all_segments)}") print(f"Combined segments saved to: {combined_file}") # Show top segments by hook strength top_segments = sorted(all_segments, key=lambda x: x.get('hook_strength', 0), reverse=True)[:5] print("\n🔥 Top 5 segments by hook strength:") for i, seg in enumerate(top_segments, 1): duration = seg['end_time'] - seg['start_time'] print(f"{i}. {seg['suggested_title']} (hook: {seg['hook_strength']}/10, {duration:.0f}s)") print(f" Episode: {seg['episode_file']}") if __name__ == "__main__": main()