Sparrow-2
A turn-taking model that knows when a voice agent should speak
Emerged finds new products worth a look, daily — what’s new today
The Duck Pond
A pond of mallards whose every quack is modelled live, not played back
A small proof that physical-modelling audio tuned by an LLM against real recordings can beat samples, and it makes good background noise.
The Duck Pond is a browser scene of hens, drakes and two broods of ducklings, placed around you with distance rings at 5 to 40 metres. Nothing is a recording: each call is synthesised as it happens from a model of the duck's syrinx and vocal tract, in Rust compiled to WebAssembly. The maker tuned the parameters by having an LLM match spectrograms to a real duck. Throw peas with P, switch to headphones with H. Free, no account; no source or maker named on the page.
More like this
How Random Are You?
Call out numbers as fast as you can while a Markov model guesses the next one
Yogurt
Meeting notes that stay on your computer, with no bot joining the call
Voxel
An open-source desktop voice changer that feeds any app through a virtual mic
Fugleramme
A Pi-powered frame that shows each bird it hears as antique natural-history art
FireRedAudio
One open 9B audio model for speech recognition, understanding, TTS and editing