Erm: CLI that strips filler words from audio recordings

Doug Calobrisi released Erm, an open‑source command‑line tool that strips filler words—“um”, “uh”, “erm”—from audio recordings. The project is hosted on GitHub and distributed as a single binary that runs on Linux and macOS [doug.sh].

Erm works by scanning the waveform for short pauses that match a configurable list of filler tokens and excising them, then re‑encoding the result. The tool accepts common audio formats (WAV, MP3) and can be chained in a pipeline, for example ffmpeg -i input.wav -f wav - | erm - -o output.wav. Because it runs locally, developers avoid sending raw speech data to third‑party services.

The utility targets developers building speech‑enabled products—voice assistants, podcasts, automated narration—where manual editing of filler words is a bottleneck. In a typical production workflow, an hour of raw speech may require dozens of manual cuts; Erm automates that step, freeing engineers to focus on higher‑level tasks such as model training or UI design. The author notes that the binary processes a 30‑minute file in under a minute on a standard laptop, making it suitable for both batch jobs and interactive use [doug.sh].

By publishing Erm under an MIT license, Calobrisi invites community contributions and encourages reuse in larger audio‑processing frameworks. The project demonstrates that even narrow‑focus tools can have outsized impact on developer productivity when they solve a repetitive pain point.