ecoSignalLab
Environmental audio is different.
It can be 24 hours long.
Or 10 years of field recordings.
Or multichannel.
Or Ambisonic.
Or messy, calibrated, industrial, spatial, fragmented, and way too large for “just open it in an editor.”
That’s why I’ve been building ecoSignalLab.
ecoSignalLab is an open-source acoustic analytics toolkit for turning large environmental, architectural, and industrial audio recordings into reproducible, inspectable, ML-ready knowledge.
It can help answer questions like:
- What happened in this recording?
- Where are the unusual moments?
- Which day sounds like this day?
- Is this archive drifting over time?
- What acoustic features are ready for downstream ML?
- Can we analyze long multichannel recordings without pretending they are small files?
The CLI is called `esl`, and the goal is simple:
Make serious acoustic analysis more practical, more transparent, and less locked inside black-box tooling.
It already supports workflows for:
- calibration-aware metrics
- long-file and shard-based analysis
- novelty and anomaly detection
- similarity search across recordings
- moments extraction
- ML-ready feature exports
- multichannel and spatial metadata
- reproducible JSON/CSV/Parquet/HDF5 outputs
This started from a frustration: We collect more sound than we can listen to.
So we need tools that help researchers, builders, architects, ecologists, and industrial teams move from “I have a mountain of audio” to “I know what changed, where it happened, and what to inspect next.”
If you work with field recordings, architectural acoustics, environmental monitoring, bioacoustics, industrial sound, spatial audio, or ML for audio, I’d love your feedback.
Repo: https://lnkd.in/eFd4G9Nt
Open-source, MIT licensed, and very much being built in public.
