AutoChop

Automated post-production studio: dead-air stripping, faster-whisper transcription, styled social captions, jump-cut assembly & YouTube chapter studio.

Comment 2

Inspiration

Every video creator knows the most soul-draining part of content creation isn't scripting or filming—it’s the first three hours in the editing timeline.

Scrubbing through raw footage, slicing out breathing pauses, deleting 3-second awkward silences, transcribing audio word-by-word, and formatting YouTube chapters is repetitive, mechanical labor. Existing SaaS tools charge high recurring monthly subscription fees, enforce strict upload quotas, or upload unreleased video footage to third-party cloud servers.

We built AutoChop AI Studio to give creators a free, local-first, privacy-preserving AI content engine that transforms raw footage into an assembled, captioned, publish-ready master rough cut in under a minute on any modern laptop.

What It Does

AutoChop AI Studio provides a complete, 3-step automated post-production pipeline:

  1. Dead-Air Stripping & Take Slicing: Automatically analyzes audio waveforms to detect silences below a decibel threshold ( \theta_{dB} ), isolates active speech segments with safety padding, and slices numbered takes (take_001.mp4, take_002.mp4).
  2. Word-Level Subtitle Burning: Uses faster-whisper (CTranslate2) to generate timestamp-accurate transcripts and hardcodes high-contrast, resolution-scaled ASS subtitles (e.g., TikTok Yellow Box, Clean Minimalist) directly onto the video.
  3. Master Rough-Cut Assembly: Stitches all non-silent takes into a seamless jump-cut master video with zero A/V sync drift.
  4. AI Metadata & Chapter Engine: Extracts spoken content to generate 5 high-CTR titles, 2-sentence description copy, YouTube-compliant timestamp chapters (strictly formatted with ( \text{start} = \text{00:00} ), ( \ge 3 ) chapters, each ( \ge 10\text{s} )), and SEO tags using any AI provider (OpenAI, Gemini, Claude, Groq, Grok, NVIDIA NIM) or a 100% offline heuristic fallback.
  5. Production Bundle Export: Packages all takes, SRT subtitles, and an edit manifest in Markdown into a one-click .zip bundle.

How We Built It

AutoChop is engineered as a modular, high-throughput media pipeline:

  • Audio Extraction & Silence Detection: Uses FFmpeg's silencedetect filter with a 2-pass audio normalization strategy. We invert silence timestamps into active speech intervals: $$I_{\text{active}} = [t_{\text{silence_end}} - \delta_{\text{pad}}, \; t_{\text{silence_start}} + \delta_{\text{pad}}]$$ where ( \delta_{\text{pad}} = 0.15\text{s} ) ensures consonant sounds like 'p' and 't' are never clipped.
  • Local Speech-to-Text (faster-whisper): Powered by CTranslate2 int8 quantization and Silero VAD filtering, yielding up to 4x faster transcription than standard Whisper while running entirely offline on CPU or Apple Silicon.
  • Media Slicing & Assembly: Slices video streams with re-encoded keyframe boundaries and stitches takes using FFmpeg's stream copy concat demuxer (with automatic filter_complex re-encoding fallback if container formats vary).
  • Styled Subtitle Engine: Generates Advanced SubStation Alpha (.ass) subtitle files with dynamic PlayResX/PlayResY coordinate scaling so captions appear crisp on vertical 9:16 Shorts as well as 16:9 widescreen videos.
  • Multi-Provider AI Studio: Features a universal OpenAI-compatible adapter supporting NVIDIA NIM, OpenAI, Groq, Grok, Claude, Gemini, and OpenCode, with client-side credential encryption and zero-key heuristic fallback.
  • Interactive UI: Built with Gradio Blocks featuring a tailored studio design system, audio waveform analysis, and take inspector.

Challenges We Faced

  1. Audio/Video Sync Drift in Jump Cuts: When concatenating multiple high-frame-rate cuts, slight discrepancies between audio sample rates and video keyframes compound over time, resulting in lip-sync drift. We resolved this by explicitly normalizing audio to 44.1kHz AAC and forcing PTS (Presentation Time Stamp) resets (setpts=PTS-STARTPTS) during slice extraction.

  2. Accidental Word Clipping vs. Silence Removal: Naive silence detectors clip speech onset or tail consonants. We designed an adaptive boundary padding algorithm combined with a minimum-gap merge heuristic: any silence under 0.2s is treated as a natural breathing pause and preserved.

  3. Gradio Multi-Provider Key Persistence: Storing and validating API keys across different providers without exposing keys in Git or public repositories required implementing an encrypted environment persistence layer with immediate visual status feedback and strict .gitignore isolation.

Accomplishments That We're Proud Of

  • Blazing Fast Local Execution: A 2-minute raw video trims, transcribes, burns subtitles, and assembles in under 25 seconds on a standard MacBook Air.
  • 100% Offline Capability: Core video clipping, subtitle burning, and heuristic chapter generation function without internet access or paid API keys.
  • Zero Hallucination YouTube Chapters: Guaranteed mathematical compliance with YouTube's strict chapter formatting rules.
  • Enterprise Test Suite: 19 automated unit and integration tests verifying silence inversion, A/V duration matching, and ASS subtitle formatting.

What We Learned

  • How to squeeze maximum performance out of FFmpeg filter graphs and subprocess pipes.
  • The mechanics of ASS (Advanced SubStation Alpha) subtitle rendering matrices across disparate video aspect ratios.
  • How to design resilient AI pipelines that gracefully degrade to offline heuristics when external APIs are unavailable.

What's Next for AutoChop AI Studio

  • Multi-Cam Auto-Switching: Automatically switch video angles based on who is actively speaking.
  • Auto-B-Roll Insertion: Use vision-language models to find points in the video where relevant stock footage or overlays can be automatically spliced.
  • Vertical Smart-Reframe: AI face tracking to automatically re-frame 16:9 landscape videos into 9:16 TikTok and YouTube Shorts.

Built With

  • ai-content-engine
  • audio-processing
  • c-translate2
  • claude-api
  • content-creation
  • faster-whisper
  • ffmpeg
  • gemini-api
  • generative-ai
  • gradio
  • groq
  • jump-cut
  • local-ai
  • natural-language-processing
  • nvidia-nim
  • openai
  • python
  • pytorch
  • silence-removal
  • speech-to-text
  • srt
  • subtitles
  • video-editing
  • youtube-automation
Share this project:

Updates

Submission history