Devising an algorithm to drop from a huge pool of automatically generated sounds those that have a pretty much static and non-modulated spectrum.
Arrived at one that takes relatively large time windows, calculates some dozen mel filter coefficients, for each calculates the ratio of 85 to 15 percentile across time, and thresholds at the mean and max of those ratio sets. Works pretty well so far.
The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!