Building a Local AI-Powered Voice Assistant

Introduction

• Discusses frustrations with cloud-based voice assistants (like Alexa)
• Goal: Create a fully local voice assistant using an AI server (Terry)
• Aim to follow a multi-level setup from basic to advanced

Level 1: Basic Setup

Parts List

• Provides a parts list for basic and advanced setups
• Mentions use of a Raspberry Pi and Home Assistant

Home Assistant

• Local home automation software
• Steps to set up Home Assistant
• Integrates with voice assistant services like Rhaspy

Initial Setup

• Use of open-source tools like Rhaspy for offline voice assistance
• Installation of Whisper (STT) and Piper (TTS)
• Addition of Assist microphone for local device input
• Setting up a wake word (e.g., "Microft")

Level 2: Expanding the System

Wyoming Protocol

• Utilizes Wyoming protocol for creating remote voice assistants
• Connects Raspberry Pi as a satellite for remote voice assistance

Setting Up Wyoming Satellite

• Installation and configuration of necessary drivers
• Creating a service to run voice assistant on a Raspberry Pi
• Testing and troubleshooting voice commands

Level 3: Advanced AI Integration

Using LLaMA (LLM)

• Integrating the LLaMA model to enhance AI interaction
• Hosting LLaMA on local hardware for faster processing

Offloading Whisper and Piper

• Moving Whisper (STT) to a more powerful server
• Similarly hosting Piper (TTS) for better performance

Enhancements

• Utilizes Docker containers for managing STT and TTS services
• Testing speed and accuracy improvements with offloaded processing

Additional Features and Troubleshooting

Custom Wake Word

• Guide on creating a custom wake word using Google Colab
• Implementation using Samba for file transfer

Custom Voice

• Attempt to create a custom voice for Terry
• Challenges faced in configuring a unique voice

Conclusion

• Successful creation of a local voice assistant
• Offloaded AI processing improves speed and functionality
• Future goals to further customize wake words and voices

Final Thoughts

• Enthusiasm for local AI solutions despite existing challenges
• Importance of privacy and control with local processing

Tools and Resources

• Links provided for various levels of setup and configuration
• Encouragement to explore further customizations and integrations