Voice Feature - Architecture

This article outlines the streamlined architecture of a voice feature that integrates Microsoft Speech-to-Text services, emphasizing secure and efficient communication without introducing new protocols or security requirements.

System Architecture

• User Interaction: The user speaks into the microphone, and the browser records the audio natively.

• Communication Flow:

  • The browser transmits the audio data via HTTPS and WebSocket to the backend.

  • This process uses existing protocols, ensuring no new communication patterns are introduced.

  • Websockets over Https are already in use in the browser for streaming the LLM answers and are no new requirements.

• Backend Integration:

  • The backend connects to Microsoft Speech-to-Text services for transcription. It connects through workload identity as the other Microsoft Services do like OpenAI.

  • The service is deployed internally within the Azure subscription, maintaining security and efficiency.

• Security and Protocols:

  • No new ports need to be opened, and security requirements remain unchanged.

  • WebSocket is already utilized for communication with large language models, ensuring consistency.

Whats new:

• deployment of Text To Speech of Microsoft.

• Inside of Uniques backend a new Microservice handles the communication.
  
  

What not changing:

• No new ports or cummunication patterns than we already use.

• No internal services of Microsoft are directly exposed.