Sort: Trending
Part 1/3:
Replacing Alexa with a Fully Local AI-Powered Voice Assistant
Dissatisfaction with Cloud-Based Assistants
The narrator expresses frustration with Alexa, describing it as "kind of dumb." They want to create their own fully local AI-powered voice assistant, tired of relying on cloud-based services like Amazon's. The goal is to connect the assistant to their own local AI server, "Terry," and have it respond to the name "Terry" with a custom voice.
Building a Local Voice Assistant with Home Assistant
The narrator starts by installing Home Assistant, a local home automation software, on a Raspberry Pi. This provides the foundation for building a local voice assistant. They then explore options for replacing Alexa's voice capabilities, finding an open-source tool called Rhasspy that enables offline voice recognition and text-to-speech.
Setting up the Voice Pipeline
The narrator walks through the process of configuring Home Assistant's voice assistant feat
[...]
Part 2/3:
Installing the Whisper speech-to-text and Piper text-to-speech add-ons.
Configuring the voice assistant to use these local services.
Testing the voice assistant's ability to control smart home devices like lights.
Expanding to Multiple Devices with the Wyoming Protocol
To create a more robust voice assistant system, the narrator introduces the Wyoming protocol, a peer-to-peer protocol for voice assistants. This allows them to set up additional "Wyoming satellite" devices, such as Raspberry Pis, that can communicate with the central Home Assistant system.
Integrating a Local Language Model (LLM)
Dissatisfied with the basic conversational abilities of the Home Assistant voice assistant, the narrator decides to integrate a more advanced language model. They set up an instance of the Llama 3.2 LLM on a separate laptop and configure Home Assistant to use it for the conversation agent.
Offloading Processing to Dedicated Hardware
[...]
Part 3/3:
To improve performance, the narrator offloads the speech-to-text and text-to-speech processing to separate Docker containers running on the laptop, rather than the Raspberry Pi. This allows them to take advantage of the laptop's more powerful hardware.
Customizing the Wake Word and Voice
The narrator attempts to customize the wake word to "Terry" and create a custom voice for the assistant, but faces challenges. They plan to address this in a future video, determined to make the voice assistant truly their own.
Overall, the narrator demonstrates a deep dive into building a fully local, AI-powered voice assistant, leveraging open-source tools and hardware to create a more private and customizable alternative to cloud-based assistants like Alexa.