"Always-On" Deepseek AI Assistant: Building an Intelligent Voice Interaction System Based on Deepseek-V3

General Introduction

Always-On AI Assistant is an innovative AI assistant project that creates a powerful and permanently online AI assistant system by integrating advanced technologies such as Deepseek-V3, RealtimeSTT and Typer. The project is especially optimized for engineering development scenarios, providing a complete voice interaction interface and command execution framework. The system adopts a modularized design and contains a basic assistant chat interface and an advanced Typer assistant session command system, which supports real-time speech recognition and text-to-speech functions. By integrating ElevenLabs' speech synthesis technology and RealtimeSTT's real-time speech recognition capabilities, the project provides developers with a complete AI assistant development paradigm, making the creation of intelligent voice assistants easier and more efficient.

Function List

• Real-time speech recognition and response system
• Intelligent dialog engine based on Deepseek-V3
• Customizable Typer command execution framework
• Multi-mode operation support (default, execution, memoryless execution)
• Dynamic memory management system (Scratchpad)
• Highly configurable assistant architecture
• Native speech recognition support
• ElevenLabs High Quality Speech Synthesis Integration
• Extensible command template system
• Real-time interactive session capabilities

Using Help

1. Environmental configuration

1.1 Basic configuration

• First clone the project locally
• Copy the environment configuration file: execute cp .env.sample .env
• Update the API key:
  • Set DEEPSEEK_API_KEY (for AI model access)
  • Setting ELEVEN_API_KEY (for speech synthesis)
• fulfillment uv sync synchronization dependency
• Optional: Install Python 3.11 (using the command uv python install 3.11)

1.2 System requirements

• Python 3.11 or later.
• Stable network connection
• Microphone equipment (for voice input)
• audio output device (computer)

2. Description of the use of the main functions

2.1 Basic Assistant Chat Interface

• Start command:uv run python main_base_assistant.py chat
• This opens a basic dialog interface
• Direct text or voice interaction is possible
• Voice response using native TTS

2.2 Typer Assistant Session Command System

• Start command:

uv run python main_typer_assistant.py awaken --typer-file commands/template.py --scratchpad scratchpad.md --mode execute

• Parameter Description:
  • --typer-file: specifies the command template file location
  • --scratchpad: set the assistant's dynamic memory file
  • --mode: set the run mode (default/execute/execute-no-scratch)

2.3 Interacting with assistants

• Clearly pronounced "Ada" wake-up call.
• Say the command, e.g., "Ada, ping the server wait for a response."
• The assistant recognizes speech in real time and executes commands accordingly
• The execution results are recorded in the scratchpad.md file

3. Description of architectural components

3.1 Typer Assistant Architecture

• Brain: using Deepseek V3 as the core AI engine
• Task handling: defined via prompts/typer-commands.xml
• Dynamic memory: state management using scratchpad.txt
• Speech Recognition: Real-time Speech to Text using RealtimeSTT
• Speech Synthesis: Integration with ElevenLabs Provides Natural Speech Output

3.2 Basic assistant architecture

• Core engine: using ollama:phi4
• Simplified design: no additional hints or dynamic memory required
• Speech recognition: also using RealtimeSTT
• Voice output: using the local TTS system

4. Advanced feature configuration

• Assistant configuration can be customized through the assistant_config.yml file
• Support for adding custom Typer commands
• Speech recognition and synthesis parameters can be adjusted
• Supports expansion of new functional modules