Hey everyone!
We’re back with an experimental build that brings some major speed improvements by introducing out-of-process Python modules! If you’re self-hosting speech-to-text (STT) and text-to-speech (TTS), expect more parallelism and faster performance!🔹 Faster & Smarter Speech Processing
WhisperLive, Coqui XTTS, F5-TTS, and Kokoro now run out-of-process, reducing bottlenecks and boosting speed.
WhisperLive can now run on a remote server, giving you more flexibility in your setup.
🔹 Action Inference System Upgrades
Flattened activated actions – organize actions better and dynamically merge child actions based on conditions.
New beforeSelectActionInference event – tweak upcoming action inference in scripts, giving you more control over actions.
🔹 UI & Experience Improvements
Pick your audio input/output device directly in the browser.
Simplified inspector with collapsible sections for easier navigation.
Italicized roleplay text for a cleaner avatar view.
🔹 Stability & Fixes
Llama.cpp improvements to prevent decoding errors in web searches.
Coqui DeepSpeed disabled by default (waiting for Python 3.12 compatibility).
Fixed multiple install issues (LlamaSharp, ffmpeg, and more).
NAudio fixes – selected audio devices will now actually be used.
Better scenario & character handling to avoid crashes and missing resources.
This build hasn’t been fully tested on Linux yet, so tread carefully! 🛠️
As always, your feedback is invaluable—try it out and let us know what you think!
How to install Voxta server app: https://youtu.be/1I9VkJ8tTloHow to update Voxta server app: https://youtu.be/5aa7sducwoc