VoiceBox

VoiceBox: A Unified AI Services Framework for Unity

VoiceBox is a flexible and extensible framework for integrating various AI services into your Unity projects. It provides a unified interface for interacting with different AI APIs, allowing you to easily switch between services and add new ones.

Features

• Unified Service Interfaces: Common interfaces for Chat, Speech-to-Text (STT), and Text-to-Speech (TTS) services.
• Service Agnostic: Easily switch between different AI service providers without changing your core application logic.
• Configuration via ScriptableObjects: Easily configure model selections, service endpoints, audio devices, and other settings in the Unity Editor.
• Audio Streaming Support: Built-in support for streaming audio for TTS, reducing latency and improving user experience.
• Extensible: Designed to be easily extended with new AI services and functionalities.
• Modding Support: Built with game modding applications in mind, providing utilities to make mod creation just as seamless as inside the Unity editor.

Natively Supported Services

• Chat

  • Google Gemini
  • ChatGPT
  • Anthropic
  • Ollama
  • Deepseek (via Ollama)

• Speech to Text

  • Azure Speech services
  • Elevenlabs Scribe
  • Whisper via WhisperLive

• Text to Speech

  • Elevenlabs

Docs

You can find the docs here: https://voicebox.timshaw.dev/

v0.3.7

• Fix a websocket issue and remove verbose logging in Elevenlabs STT

v0.3.6

• Fix an issue where Elevenlabs STT query parameters would serialize incorrectly in some locales (Thanks @xCore!)

v0.3.5

• Fix an issue with NAudio not detecting devices with names over 31 characters long

v0.3.4

• Fix a WebSocket error when attempting to reconnect to a TTS service

v0.3.2

STT

• Update AzureSTTServiceManager to provide VoiceBoxResultReason.RecognizedSpeechWithTimestamps when requestWordLevelTimestamps = true.
• Update AzureSTTServiceManager to save derived config.

TTS

• Update ElevenlabsTTSServiceManager to hide event not used warnings.

v0.3.0

This update contains breaking changes!

TTS

• Rework interface
  • Add voice cloning API call support
• Implement more elevenlabs TTS api config options
• Correctly set model ID in elevenlabs requests

STT

• Add Elevenlabs STT support
• Implement local VAD for STT to limit silent chunks sent to STT api
• Add support for word-level timestamps

Misc

• Resolve several issues with service cancellations
• Rework audio decoder class to support wave audio
• Rework audio streaming to stream through an audioclip, allowing for filters to be added to the output

v0.2.0

Initial Thunderstore release