README.md

metadata

title: RSR AI
emoji: 🔊
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.30.0
python_version: '3.12'
app_file: app.py
pinned: false
short_description: Enhance robotic-sounding video audio
startup_duration_timeout: 30m
license: apache-2.0

RSR AI — Robotic Sounds Remover AI

RSR AI is a video-audio enhancement Space. Upload a video, enhance its audio with an open-source speech-enhancement model, and download a new video containing the enhanced audio and the original video stream.

Pipeline

• Speech enhancement: ClearVoice with MossFormer2_SE_48K (Apache-2.0 model), which is designed for full-band speech enhancement and noise reduction.
• Media I/O: FFmpeg / FFprobe extract and replace the audio stream.
• Video preservation: the video stream is remuxed using stream copy (-c:v copy), not decoded and re-encoded. The app does not generate or modify video frames.
• Default settings: 100% enhancement strength, Balanced preset, preserve original video timing enabled.

Limitations

MossFormer2_SE_48K is a general speech-enhancement model, not a dedicated anti-vocoder or robotic-voice reversal model. It may improve speech clarity and noise/artifacts, but it cannot reliably reconstruct information that is absent from the source audio. It may also change music, sound effects, or speech. Results depend on the video. Keep your original file.

The app limits processing to 180 seconds per video to avoid exhausting free CPU Space resources. Model weights are fetched on first use, so the first run can take longer and requires sufficient disk and memory.

Run locally

Use Python 3.12 and install FFmpeg/FFprobe, then install packages from requirements.txt. Launch with:

python app.py

Deploy to Hugging Face Spaces

1. Create a new Space at https://huggingface.co/new-space.
2. Choose Gradio and use the repository name rsr-ai (the display title is RSR AI).
3. Upload app.py, requirements.txt, packages.txt, and this README.md to the root of the Space repository.
4. Wait for the build to finish. The model weights download the first time you press Enhance audio.

The bundled files are a deploy-ready project, but they do not create or publish a Space in your Hugging Face account automatically.