Clean voice recordings with AI
Reduce noise and uneven speech levels, then compare the result with the original. Optional filler removal produces a cut list that needs listening and review.
Clean voice recordings with AI while keeping listening central to the decision. The voice-cleaner skill uses local tools to reduce mains hum, background noise and uneven speech levels, then normalises towards a chosen loudness target. It does not provide complete studio-voice reconstruction or every feature of an audio production service.
Optional filler and short-repeat removal uses locally transcribed word timestamps. Missed words can produce inappropriate cuts, so retain the cut list and listen around each edit. When the input is video, the shortened output is audio only and no longer automatically matches the picture.
What it does
- Reduce noise in spoken-word recordings.
- Balance uneven speech levels.
- Compare original and processed WAV files.
- Inspect metrics alongside a spectrogram.
- Review an optional word-cut list.
- Choose an appropriate audio output format.
Install and use
- Provide the complete package to a file- and command-capable agent with authorised audio. Prepare FFmpeg 6 or newer, ffprobe and Python 3.10 or newer with NumPy. Optional filler cutting needs faster-whisper and an initial model download; the local route needs no cloud account.
- For example, request noise reduction and level balancing for your own recording, preserving the original and producing comparison files without removing words.
- Listen to both versions and inspect the measurements. If you request cuts, review their locations and separately restore video synchronisation where needed.
- Treat the optional web route separately: configure Cloudflare Access before sharing and protect the VPS. Job files persist until the operator schedules cleanup; hosting costs are separate.
FAQ
Source
Made by Project Alpha Tech. Original skill: voice-cleaner, published on GitHub under the MIT licence.
View on GitHub