The basic idea
An acoustic index is a mathematical summary of an audio recording designed to capture some aspect of the underlying ecosystem. Examples: the acoustic complexity index (ACI) measures temporal variation in acoustic energy across frequency bands; the acoustic diversity index (ADI) measures how acoustic energy is distributed across frequency bands; the bioacoustic index measures the ratio of biological to anthropogenic acoustic energy. Each index reduces an audio recording — potentially hours of complex sound — to a single number or a small vector of numbers. The numbers can then be compared across sites, seasons, or years to track ecosystem-level changes.
Different ecosystems sound different.
Why it's plausible
Different ecosystems sound different. A healthy old-growth forest at dawn produces a richer, more layered soundscape than a degraded agricultural area at the same hour. The richness has acoustic signatures: more species vocalizing, more distinct frequency bands occupied, more temporal variation as different species cycle through dawn-chorus prominence. If those signatures are reliable enough across enough habitats, summary indices computed from the audio should correlate with traditional biodiversity measures — and they often do, with caveats. Bryan Pijanowski's research at Purdue helped establish the framework in the 2000s and 2010s, and the field has developed substantially since.
What works
Acoustic indices generally correlate with biodiversity measures across coarse comparisons — degraded versus pristine habitats, urban versus wild sites, drought versus wet years in semi-arid ecosystems. For tracking large-scale changes in ecosystem state, acoustic indices have been validated repeatedly. They're cheap to compute, they don't require species identification, they aggregate naturally across long recordings, and they produce comparable numbers across sites. For applications where the question is 'is this habitat healthier or less healthy than that one' or 'has this site changed across the past decade,' the indices work.
What doesn't work as well
Acoustic indices have struggled with finer-grained questions. Two sites with the same overall acoustic complexity can have very different species compositions. Acoustic anomalies (cicada emergence years, frog breeding events, single-species concert effects) can swamp underlying biodiversity signals. Anthropogenic noise (traffic, agriculture, distant industry) confounds many indices in ways that are hard to correct for. Index values can be similar across habitats for different reasons, which means index comparisons aren't always meaningful even when they look meaningful. The cleanest published critiques of acoustic indices (multiple papers in the late 2010s and early 2020s) caution against using single-index summaries for high-stakes conservation decisions without ground-truth validation.
What's happening now
The 2020s have seen a partial shift away from single-index summaries and toward learned acoustic embeddings from foundation models as the new summary representation. A BirdNET[2] or averaged over a recording produces a vector that captures more of the soundscape's compositional information than any single hand-designed index. The trade-off is interpretability — a 1,024-dim vector means something to a downstream classifier but doesn't have the 'plain English' summary value that ACI or ADI did. The field is converging on a hybrid practice: foundation-model embeddings for compositional questions, classical indices for trend-tracking and gross-comparison work.
The implication for AI bioacoustics
Acoustic indices and contemporary AI bioacoustics aren't competing approaches; they're complementary. The indices ask 'how does this soundscape compare?' The AI methods ask 'what's in this soundscape?' For ecosystem monitoring at scale — passive acoustic monitoring deployments across a national forest, multi-site comparative studies, multi-decade trend tracking — both approaches contribute. CrowLingo's project doesn't directly use acoustic indices because we're focused on a single species and questions about its vocal repertoire, but the broader acoustic-monitoring landscape that funds and motivates bioacoustic research relies heavily on indices, and the AI methods our work uses came partly from the same intellectual tradition.