Skip to content

Ratings and global indices

Sessions are reportable in three established idioms beyond the corpus's own descriptor set: HVAC/room criteria, environmental-noise descriptors, and soundscape-ecology indices. There are also spatial descriptors that only an ambisonic corpus can supply. The borrowed idioms earn their place because a building-services engineer, a noise consultant and a soundscape ecologist each want the same recording expressed in the numbers their own field argues in. Everything except the room criteria is appended automatically to the analyze summary and README table.

Room noise criteria (ventilation & building services)

from ambiscape import iso

spec = iso.background_octaves_db(F, pct=50, offset_db=cal["dbfs_to_dbspl"])
iso.room_criteria(spec)
# {'NR': 28.3, 'NR_governing_hz': 250, 'NC': 27.9, 'RC': 25.6,
#  'RC_class': 'R'}

NR (ISO/R 1996, analytic curves), NC (ANSI S12.2, tabulated), and a simplified RC (Blazier) with rumble/hiss classification, which together are the language HVAC noise is specified in worldwide. They mean something in absolute terms only with SPL calibration (calibration.json: dbfs_to_dbspl); uncalibrated ratings compare rooms recorded with one recorder and gain setup only. Compute them per machine state (states.state_segments masks) to rate "vent on" vs "vent off" separately.

Environmental-noise descriptors

  • Intermittency ratio IR (analysis.intermittency_ratio, Wunderli et al. 2016): the share of energy carried by events, near 0 for steady drones and dense traffic, high when energy arrives as distinct events. It is the single best complement to events/min.
  • Emergence LAeq − LA90: how far the energetic mean sits above the background, which is the classic "how eventful" number. Being built on an energy average it is set by the loudest frames of the span, so in a quiet room it should be read next to the trimmed level and the percentiles—see Reading energy averages.

Prominent tones (ambiscape tones)

DIN 45681-style tone detection on the cached per-minute mean spectra: each narrowband spectral peak is compared against the masking-noise level of its surrounding critical band, giving a decibel prominence ΔL = L_tone − L_noise per tone (≥ 6 dB, the decisive audibility criterion of DIN 45681, counts as prominent). Detections are aggregated over time, so a ventilation or appliance hum shows up as one persistent line with its presence fraction, while a passing siren does not:

ambiscape tones SESSION/          # needs a prior analyze run
#   152.3 Hz: ΔL 35.5 dB (max 35.5), present 100% (180 min)

The strongest persistent tone lands in the analyze summary and README as tonal_prominence_db / tonal_prominence_hz (with n_prominent_tones), per state in the state-resolved table — rate "vent on" vs "vent off" tones separately — and the percentile-LTAS figure marks the top tones. ΔL is a level difference, so it is meaningful without SPL calibration. This follows the method of DIN 45681, not a certified implementation (no frequency-dependent masking index or uncertainty term); pure numpy/scipy, always available.

The summary also carries fluctuation_index, a broadband 4 Hz-weighted modulation-depth index from the cached 20 ms envelope (≈ 0 for a steady drone, high for a ~4 Hz wobble) — the cheap companion to the approximate iso.fluctuation_strength (vacil) computed per ear by ambiscape iso (see the room-acoustics page).

Ecoacoustic indices (ambiscape.ecology)

ACI, ADI/AEI, NDSI, BI, and acoustic entropy H make up the standard soundscape-ecology battery, computed from the cached 96-band spectrogram. They are reported for cross-corpus comparability. Indoors, read NDSI/BI as "energy in the 2–8 kHz band", not as proof of biophony, since a ventilation hiss scores as birds. Combine with the taxonomy layer before interpreting.

ACI is averaged over complete 5-minute chunks, and its magnitude depends on the chunk length, so a recording shorter than one chunk has no comparable value: aci is then None. Clip corpora of 5–30 s carry no ACI at all.

Spatial descriptors (ambiscape.spatial)

  • Directional entropy: how many directions the place sounds from (0 = one bearing, 1 = even around the horizon), the spatial analogue of a diversity index.
  • Horizon fractions: energy from above / around / below ±10° elevation, which separates ceiling-mounted services and birds from footsteps and ground traffic.
  • Foreground/background azimuth overlap: Bhattacharyya overlap of the loudest-25 % and quietest-25 % direction histograms, 1 when figure and ground share a direction (one-source rooms), 0 when they occupy different sectors.

Room-acoustics companions

analysis.decay_metrics extends decay_time (unchanged, its numbers feed frozen reports) with EDT, C50/C80, and D50 per octave from the same truncated-Schroeder decay, which give reverberance and clarity from any good impulse in the recording, a door slam or a dropped book.