Freesound's content-based similarity search returns sounds that are acoustically alike, not just similarly tagged

Beyond text search, the Freesound APIv2 offers content-based audio similarity search: given a reference sound (by ID), it returns the most acoustically similar sounds from the whole database, measured on descriptors — spectral features, rhythm, timbre — extracted by Essentia (the open-source analysis library from the MTG/UPF). This differs fundamentally from tag search: a query for ‘kick drum’ returns anything tagged that way, whereas a similarity search from a specific kick returns sounds that perceptually sound like it even if untagged. It enables ‘find more like this’ and sound-replacement workflows: a performer can keep the feel of a palette while continuously swapping in fresh material — crate-digging for sounds with the same energy. Related tooling exposes this spatially: ‘Freesound Explorer’ places sounds in a 2D space by estimated acoustic properties for visual browsing, and samplers like SOURCE hit the /sounds/<id>/similar endpoint to swap a loaded sound for an acoustically similar one.

Examples

A live-coder loads a metallic clang and requests ‘find similar’; the API returns acoustically similar sounds (door bells, brake squeals, metal percussion) and swaps one in for the original. Freesound Explorer displays results spatially — closer points are more acoustically similar. SOURCE replaces loaded sounds via Freesound’s similarity feature during a live set.

Assessment

Explain the difference between a text-tag search and an audio-similarity search on Freesound, and what ‘acoustically similar’ means versus ‘similarly tagged.’ Describe a live-performance scenario where similarity search beats keyword search, and how you would use it in a sampler like SOURCE.