Voxta 143 Beta: Real-Time Audio Streaming + 2 New TTS: CSM & ORPHEUS!

Hey everyone!

Big news! We're excited to release Voxta Server v1.0.0-beta.143! The star of this update is real-time TTS streaming – hear your characters speak much sooner as the audio generates! Experience more natural conversations as you no longer have to wait for the full audio clip to finish processing.

We're also introducing two powerful new experimental AI voices: Canopy Labs Orpheus and Sesame CSM.

🔊 Streaming Audio & New Voices

• Real-Time TTS Streaming: Audio starts playing almost instantly for supported TTS (F5, Orpheus, Kokoro, CSM, Voxta Cloud, NovelAI, OpenAI, ElevenLabs).

• Canopy Labs Orpheus TTS: A new voice engine supporting tags like <laugh>, <sigh>, <sniffle>, <groan> and <sigh>.

• Sesame CSM TTS: Another new experimental voice service. The base model isn't super impressive yet, but we're expecting much better results with upcoming fine-tuned variants. Still, it's fun to try out — like this one, which can do whispery ASMR-style speech.

💬 Chat & Interaction Upgrades

• New Chat Commands: /regenerate the last character message or /rollback recent conversation history with ease.

• LlamaCpp Updated: Now running the latest version (0.23.0) for improved performance and features.

⚙️ Core Improvements & Stability

• Python Services Reliability: Significant improvements to prevent background Python processes from stalling.

• F5-TTS Flexibility: Added support for older pre-1.0 multilingual models.

• Better Organization: Place voice samples in subfolders and configure custom paths for some models.

• Updated Foundation: Moved to Python 3.12.9 & Torch 2.6. (Requires reinstalling Python packages - see notes).

🖥️ Voxta UI Enhancements

• Multiple Memory Books: Assign multiple memory books to characters and scenarios. Organize memory into separate books (e.g., lore, items, plot etc.) and mix & match as needed, instead of cramming everything into one.

• Key Fixes: Addressed issues with Firefox audio recording, attachment display/sending, preset page settings, profile saving, and more for a smoother experience.

• Better Error Handling: More robust error catching to prevent unexpected issues.

Important Notes: 🛠️

• Real-time TTS streaming and the new Orpheus/CSM AI voices can be demanding on your GPU, especially running alongside local LLMs. High-end hardware is recommended.

• Due to the Python 3.12.9 upgrade, you'll need to re-install Python packages. After updating Voxta, you can safely delete the old Data/Python/python-3.12.8-amd64 folder.

• To download CSM-1B and Llama 3.12 1B, you need approval on HuggingFace. You can then create an environment variable HF_TOKEN and Voxta will use it automatically, or you can download them manually (instructions to be provided later).

This is a beta release, so your feedback is invaluable! Give the new features a spin, especially the audio streaming and new text to speech services and let us know what you think on Discord or here!

How to install Voxta server app: https://youtu.be/1I9VkJ8tTlo
How to update Voxta server app: https://youtu.be/5aa7sducwoc