I recently released Sound Finder, a Unity Asset Store plugin, and building it meant solving an unusual problem: running up to twelve neural networks inside the Unity editor, without Sentis, ONNX Runtime or Python. The plugin searches a sound library by how things sound. That takes models that turn sounds and text into fingerprints you can compare, a pitch model, an onset detector, and two text-to-audio models built from four networks each. They all run on an inference runtime I wrote in C# and H
I recently released Sound Finder, a Unity Asset Store plugin, and building it meant solving an unusual problem: running up to twelve neural networks inside the Unity editor, without Sentis, ONNX Runtime or Python. The plugin searches a sound library by how things sound. That takes models that turn sounds and text into fingerprints you can compare, a pitch model, an onset detector, and two text-to-audio models built from four networks each. They all run on an inference runtime I wrote in C# and HThis article was originally published by Dev Community.
Read Full Story on Source Site →
