Zina - Autonomous voice assistant

Zina - Autonomous voice assistant

Purpose

The autonomous voice assistant "Zina" is designed for use in secure meeting rooms and as a device control center in a local trusted network.

Key users are employees of organizations working with confidential information, as well as users who need autonomous control of smart devices.

"Zina" allows you to record event transcripts, generate short excerpts of meeting results and create orders, ensuring complete autonomy of work without the need for an Internet connection.

Advantages

  • Completely autonomous work without an Internet connection.
  • Using advanced technologies of large language models (LLM) for natural language processing.
  • Ability to manage trusted devices in a local network.
  • Recording event transcripts and generating short excerpts.
  • Good price/quality ratio, thanks to the use of the NVIDIA RTX 3050 graphics accelerator.
  • Adaptation of the Mistral 7B language model for working with the Russian language.

Relevance

Modern organizations increasingly face the need to process confidential information in conditions of limited access to the Internet. Traditional cloud solutions are not suitable for such tasks due to the risk of data leakage. "Zina" offers a fully autonomous solution that ensures data security and allows you to effectively manage local devices, record and analyze information in real time.

Technology

"Zina" is an autonomous voice assistant built on the basis of a modified Mistral 7B language model adapted to work with the Russian language.

Zina-1

The main components of the system:

  • Speech recognition neural network: converts voice commands into text.
  • Text-to-Text neural network: provides semantic processing of information using the LLM model.
  • Speech generation neural network: provides voice feedback to the user.
  • Visualization module: a digital embodiment of "Zina" using lip sync technology.

"Zina" can be integrated into a local network to control trusted devices such as smart lamps, cameras and other IoT devices. The assistant is also capable of recording event transcripts, analyzing them and generating short excerpts, making it an indispensable tool for holding meetings and events in secure spaces.

Zina-3

Two versions of the "Zina" voice assistant have been implemented:

  • Smart speaker: has a built-in speaker.
  • TV set-top box: sound and image are output via HDMI.

The Zina project was presented at the Army-2024 exhibition and received high marks from experts.

Technology readiness level

TRL 4: A detailed mockup of the solution has been developed to demonstrate the technology's functionality