NovaVoice

NovaVoice is an AI voice assistant. Currently very early in development, contributions welcome!

Demo

demo.mp4

Features

• 🎤 Wake Word Detection (Porcupine or Azure Speech)
• 🗣️ Speech-to-Text (Built-in using WebRtcVad and Whisper Turbo or Cheetah)
• 🔊 Text-to-Speech (Windows Speech or Mimic3)
• 🤖 AI Interactions (via Groq API)
• 🔍 Extensible Tool System (Google Search API, YouTube Data API, AccuWeather API)
• 📝 Conversation History
• 🎵 Media Playback Support
• 🔄 Configurable Audio Processing Pipeline

Prerequisites

• .NET 8.0 SDK
• Required external dependencies:
  • yt-dlp (for YouTube playback)
  • mpv (for audio playback)
  • mimic3 (if using Mimic TTS)

Configuration

Configure the assistant through environment variables or appsettings.json:

{
  "Recorder": "Picovoice",
  "WakeWordProvider": "Porcupine",
  "SpeechToTextProvider": "Cheetah",
  "TextToSpeechProvider": "WindowsSpeech",
  "Google": {
    "ApiKey": "YOUR_KEY",
    "CustomSearchId": "YOUR_ID"
  },
  "AccuWeather": {
    "ApiKey": "YOUR_KEY"
  },
  "Groq": {
    "ApiKey": "YOUR_KEY",
    "WhisperModel": "whisper-large-v3-turbo",
    "GeneralModel": "llama-3.1-70b-versatile",
    "ToolUseModel": "llama3-groq-70b-8192-tool-use-preview",
    "SystemPrompt": "You are a helpful, child-friendly voice assistant called Nova. Respond concisely to user prompts (received via Automatic Speech Recognition). You have access to the following tools: `google_search`, `youtube_search`, and `check_weather`. Consider carefully when you should use them.\n\n- Use `youtube_search` to find music and audio content that the user may want to listen to.\n- Use `check_weather` to get the current weather conditions or forecasts for a location.\n- Use `google_search` for retrieving live data, for example news or exchange rates\n\nPrioritise safety and well-being, never provide responses that could be harmful, inappropriate, biased, or misleading. If the prompt is ambiguous, unsafe, or you are unable to provide a helpful response using the available tools, respond with \"Sorry, I can't help with that\"."
  },
  "Picovoice": {
    "AccessKey": "YOUR_KEY",
    "Porcupine": {
      "Keywords": ["Hey Nova", "OK Stop"]
    }
  }
}

Project Structure

• Speech/ - Core speech processing components
  • Recorder/ - Audio input handling
  • SpeechToText/ - Speech recognition implementations
  • TextToSpeech/ - Speech synthesis implementations
  • WakeWord/ - Wake word detection providers
• Tools/ - Extensible tool system implementations
• Models/ - Data models and configuration classes
• Events/ - Event handling system

Getting Started

1. Clone the repository
2. Configure your API keys in appsettings.json or environment variables
3. Install required external dependencies
4. Build and run the project:

dotnet build
dotnet run

Provider Options

Wake Word Detection

• Porcupine
• Azure Speech

Speech-to-Text

• Built-in (crude implementation using WebRtcVadSharp and Whisper Turbo)
• Cheetah

Text-to-Speech

• Windows Speech
• Mimic3

Recording Engine

• NAudio
• Picovoice

Tool System

The assistant includes several built-in tools:

• google_search - Web search functionality
• youtube_search - YouTube content search and playback
• check_weather - Weather information retrieval

Acknowledgments

• Thanks to UNIVERSFIELD for the sound effects.