ambiscape is a Python toolbox for analysing soundscapes—the sonic ambiences of rooms and other places. It reads mono, stereo, binaural, or first-order ambisonic recordings of any length and describes a place's sound as a whole: level, spectrum, space, rhythm, sources, and more. Several recorders spread through a building can be read together as one acoustic network.
pip install ambiscapeOptional extras add psychoacoustic indicators, machine listening, music analysis, live capture, and spatial visuals. See the install guide.
Point analyze at a session—a folder of WAV files from one recording occasion:
ambiscape analyze my-session/This streams the audio in constant memory, however long it is. It extracts features, computes descriptors (Leq, LAeq, percentile levels, event statistics, diffuseness, and more), renders overview figures, and writes a README.md summarising the session. Start by reading that README and looking at analysis/overview.png:
The quickstart guide continues from there, on the command line and in Python.
For a recording you keep as a folder of its own, ambiscape init my-session/ writes a session.json to fill in, and ambiscape run my-session/ runs the whole chain on it, from analyze to a filled report; see one folder per recording.
analyze is one of nearly forty subcommands. The others cover taxonomy annotation, rhythm and tonality, room acoustics and impulse responses, ecological and source-domain indices, perceptual surveys, multi-recorder building networks (network), corpus aggregation, and privacy-aware publishing. The command overview lists them all; ambiscape --help prints the same list.
- User guide & API reference—the session model, feature and descriptor definitions, and a page per analysis module.
- Wiki—field-recording protocol, recipes, worked case studies, design rationale, and research context.
ambiscape analyses sound; its sister toolbox MGT-python analyses video. The two meet at file boundaries—see Working with other packages.
Parts of the toolbox that work on their own:
- the streaming reader, which walks a recording of any length in constant memory and hands you frames
- the level and spectral feature extractors (Leq, LAeq, percentile levels, log and mel spectra) as functions on arrays
- the ambisonic direction and diffuseness estimators, which take a first-order AmbiX block and return where the energy comes from
- the session model, a folder of files on one clock with a per-take feature cache, which any batch analysis can adopt
- the ISO 12913-2 perceptual survey tools and the impulse-response and auralisation functions
- the FLAC hand-off convention, a leading
YYYYMMDD_HHMMSSstamp in local time, which the other three toolboxes read
ambiscape describes places, not music: per-track music analysis lives in musiscape, and deeper music information retrieval in librosa or the MIR Toolbox. It streams long files but is not a real-time engine; the capture extra records, it does not react. It does not read motion or video; those go to micromotion and musicalgestures.
Four packages from the fourMs lab, each released separately on PyPI. Which one you want is decided by what you have in hand rather than by what you want to know:
| you have | use | it gives you |
|---|---|---|
| a recording of a place — mono, stereo, binaural or ambisonic | ambiscape (this one) | the sonic ambience of that place: level, spectrum, space, rhythm, sources |
| a motion time series from a body — optical markers, an accelerometer, a respiration belt, a force plate | micromotion | quantity of motion, posture, balance, and the band conventions the others follow |
| a video file, with or without its sound | musicalgestures | motiongrams, videograms, motion analysis from ordinary video |
| a folder of music, or a concert recording | musiscape | many tracks and albums compared at a glance |
Where a measure appears in more than one package it has a single owner and a single implementation, so the answer does not depend on which package you called. micromotion owns filtering, lag estimation and circular statistics; ambiscape owns the soundscape descriptors; musicalgestures owns everything that starts from pixels. Music analysis moved out of ambiscape into musiscape on 2026-08-12, so a release of either from before then may still carry the other's functions.
ambiscape installs and runs without any of the others. It keeps its own copy of six short
circular-statistics primitives rather than taking a dependency for them, which is a deliberate
exception to the single-owner rule: tests/test_circstats_agreement.py checks them against
micromotion's and skips when it is not installed. That test exists because the two Rayleigh
implementations were once found to disagree on about a fifth of random cases — ambiscape used
Zar's series expansion and micromotion uses Wilkie's approximation, both published, neither
wrong, and nothing anywhere saying they were meant to match.
MIT—see LICENSE.
ambiscape is developed as part of the AMBIENT project at fourMs / RITMO, University of Oslo, supported by the Research Council of Norway. It is the streaming companion to ambiviz, which renders rich spatial visuals from short ambisonic files.
Cite the concept DOI, which always resolves to the newest version:
Jensenius, A. R., & Guo, J. (2026). ambiscape: analysis of soundscapes from mono, stereo, binaural and ambisonic recordings [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.21965234
Where the exact behaviour matters, add the version you ran. Every release has its own DOI, listed on the Zenodo record.
An older concept DOI, https://doi.org/10.5281/zenodo.21948997, is frozen at 0.42.0. It was created by a hand deposit made on 2026-08-15, before the Zenodo GitHub integration was archiving this repository; the integration began working the next day and every release since is under the DOI above. Zenodo cannot merge two concepts, so both records exist and only one of them advances. Cite the DOI above.
CITATION.cff in this repository carries the same information in machine-readable form.