The Home Assistant project, which develops an open home automation platform, announced the development of a hardware device called Home Assistant Voice with a voice assistant implementation. The device can be used to control a smart home, perform tasks, and generate answers to user questions when interacting in natural language. The circuit diagrams, board layout, and layouts for printing the case on a 3D printer are distributed under open licenses. The firmware code is available under the MIT license. The finished device is available for sale for $59.
Home Assistant Voice is optimized for use with smart home devices based on the open Home Assistant platform. To use the assistant, simply turn on the device, and Home Assistant will automatically offer to integrate it into your home network. Compared to commercial systems on the market, such as Google Mini, Apple HomePod, Amazon Echo, and Yandex Station, the Home Assistant project allows for a completely autonomous solution, independent of external services and internet connection. Speech recognition and synthesis, as well as response generation, are performed on a separate platform. server using AI technologies and a large language model.
The user can run their own server in their home network or use an external cloud service. For local processing of voice commands, it is recommended to use a system with at least an Intel N100 processor. The software environment is based on the Assist component, which uses the Whisper model or the Rhasspy Speech system for speech recognition. The large LLaMA language model is used to generate answers to questions asked in natural language. If you do not want to create your own server, the user can connect the voice assistant to the Home Assistant Cloud service. In addition, integration with the Open AI, Google AI and Anthropic AI services is possible.
The user-facing device measures 9.4 x 9.4 x 3 cm and weighs 120 grams. The device is equipped with an ESP32-S3 SoC based on a dual-core XTensa LX7 240 MHz microcontroller with 16 MB Flash and 8 MB RAM. The XMOS XU316 chip is used for sound processing, performing functions such as echo suppression, noise elimination and automatic adjustment of the input signal level. Connection to the home network and other devices is carried out via Wi-Fi and Bluetooth 5.0. A Grove port is provided for connecting additional sensors and accessories. 
At the current stage of development, a preliminary version of Home Assistant Voice PE (Preview Edition) has been released for sale. This preliminary status is due to the desire to incorporate community feedback and create a fully functional product ready for use in any home in any country. In its current form, not all planned functionality is ready, and not all languages are supported. For example, Russian is already listed as supported locally. server and in the cloud, but the Ukrainian language can currently only be used when connected to an external cloud.
With the help of the assistant, the user can use voice commands to turn on/off the light, activate household appliances, add notes to the shopping list, control the air conditioner, get information about the temperature and time, start the timer, open the gate, play music, start the robot vacuum cleaner and perform scenarios pre-defined in Home Assistant. The body is equipped with a backlight for visual indication of the reception of commands. There is a hardware microphone switch, which guarantees privacy when the user needs it. External speakers can be connected via the 3.5mm audio port to play music.
Source: opennet.ru
