Audio and Embeddings join the party
SOTA bioacoustics classification, audio playback, spectrogram visualization. Just select an audio model and select an audio file.
Better GUI experience
All visualizers got reworked. Now it's more enjoyable to navigate images, video files, live feed, and now, audio files! Live feed now includes an adjustable buffer, in case yo uwanna go back in case you missed something.
New models
Like always at: https://boquila.org/hub
You can run Perch 2, BioClip 2 and a lot more! Just import the .bq model and give it some data, it's literally that easy.
Code
- We went from drawing a lot of things by "hand" (manipulating the image buffer) to just using egui to draw things. This is much better because it allows a lot more interactivity. We are still keeping the functionality to draw directly on the image buffer but that's kept for exporting.
- Egui plots was incorporated for great spectrogram visualization
- Some code got more ugly with the purpose to finish features (audio and embeddings), this will be cleaned up. Luckily, adding a new type of output (audio classification probabilities, embeddings) is extremely easy thanks to the Enum AIOutput
What's Changed (some PR logs)
- Audio AI inference, plus rewrites on AI handling by @jdiaz97 in #30
- Small code and project cleanups by @jdiaz97 in #31
- removed some default deps and performance improvements by @jdiaz97 in #32
- deleted egui_extras by @jdiaz97 in #33
- Big rewrites, more complete audio analysis, better GUI by @jdiaz97 in #34
- support for embedding models by @jdiaz97 in #35
Full Changelog: v0.4...v0.5
