henliveira

henliveira / av-curator

Public

Audio-visual data curation pipeline — scene cuts, silence trim, dedup, CLIP/Whisper filtering for messy web video.

45
0
89% credibility
Found May 25, 2026 at 45 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

AV-Curator is a video cleaning tool that helps researchers prepare raw video clips for machine learning by automatically removing unwanted content like silent sections, black frames, duplicate videos, watermarks, and off-topic language, producing a clean dataset with a detailed report of what was kept and removed.

How It Works

1
📁 Gather your video clips

You collect all your raw video files into one folder on your computer, ready to be organized.

2
🔍 Create a catalog of your videos

The tool scans your folder and automatically creates a detailed list of every video, noting how long each one is and what format it uses.

3
⚙️ Choose your cleaning strategy

You pick a preset that matches your goal—either finding clips with clear speech for training a transcription AI, or finding clean visual clips for a video-understanding project.

4
🧹 Watch the magic happen

The tool runs through your videos one by one, checking each against your chosen criteria and automatically removing or trimming away unwanted sections like silent parts, black screens, or duplicate content.

5
📊 Review your cleaning report

You see a clear visual breakdown showing exactly how many videos made it through each stage, so you understand what was kept and what was removed at every step.

6
Decide what to do with your clean clips
📦
Export as-is

Keep your trimmed videos in their current form and move straight to using them for training your AI model.

✂️
Auto-slice clips

Let the tool automatically cut your videos to remove black frames, silence, and scene transitions, producing perfectly clean clips.

✅ Your training data is ready

You now have a clean collection of video clips, perfectly suited for training your AI model without any of the messy problems that would have hurt quality.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 45 to 45 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is av-curator?

av-curator is a Python-based pipeline that cleans up messy web video for machine learning training. It runs a configurable chain of filters to strip title cards, trim silence, remove near-duplicates, and verify that audio actually contains speech in the right language. The CLI tool `av-curate` handles manifest building, pipeline execution, and clip slicing in three commands. Built-in filters range from cheap ffprobe checks to heavyweight CLIP and Whisper inference, with automatic caching so you don't re-run GPU filters unnecessarily.

Why is it gaining traction?

The modular filter design means you can start simple (scene-cut + silence-trim) and layer on heavier checks as needed. Every filter decision is logged per clip, so you get a full audit trail of why something was kept or dropped. The funnel report shows exactly how many clips survive each stage, which is rare in data curation tooling. Presets for speech-heavy ASR training and clean video for video-language models give you working configs in seconds.

Who should use this?

ML engineers building audio-visual datasets from YouTube or other web sources. Researchers preparing training data for speech recognition, video-language models, or deepfake detection. Data engineers who need to document their curation decisions for reproducibility.

Verdict

av-curator fills a real gap in the MLOps stack, but at 45 stars it's still early-stage. The documentation is solid and the filter contract is clean, but test coverage and long-term maintenance are unknowns. Worth evaluating for production pipelines if you're comfortable with a project that may still see breaking changes.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.