Flexible visualization of sound collections
summary
A browser library is built that converts framewise spectral descriptors into per-sound vectors, constructs a K‑nearest‑neighbor graph (k≈15), and experiments with PCA, t‑SNE, UMAP and force/constraint layouts to place samples. Tested on a few thousand drum and loop samples, UMAP fast with minimum‑distance parameter gives compact clusters, constraint layout enforces non‑overlap but grids results, and force‑directed + collision works for variable sizes/shapes with some residual overlaps. Segmentation of long recordings into sample collections is implemented as an optional preprocessing step.
AI-written summary · about
chapters (5)
related talks
04mThe Sound Sketchpad: Expressively Combining Large and Diverse Audio Collections
16mA JavaScript Library for Flexible Visualization of Audio Descriptors
51mArt of Effective Visualization of Multi-dimensional Data
14mSonification: You Can See Your Complex Data with Sound!!
40mBuilding A Folk Music Recommendation System with LLMs
43mMake It Sound
38mData Mapping for Data Exploration
21mA Visual Interaction Framework for Dimensionality Reduction Based Data Exploration