Skip to main content

6.5.4 AI Chatbot

Last Version: 19/09/2025

Watch SpacemiT Official Video

Overview​

This demo showcases a fully local intelligent voice assistant system. It integrates speech recognition, natural language processing (NLP), and speech synthesis into one workflow.

All AI models are converted to the ONNX format and optimized through quantization. With the K1 chip’s hardware acceleration, the system ensures fast response while maintaining high-quality interactions.

Hardware​

Supported Development Boards​

  • Muse Book
  • Muse Pi
  • Muse Pi Pro
  • Banana Pi（K1 version）

System Requirements​

  • Operating System: Bianbu OS 2.0 or higher
  • Storage: At least 8GB available
  • Memory: 4GB or more recommended

Core Modules​

This project includes the following core modules:

  • Audio Capture Module Captures audio input via an external USB microphone.

  • ASR Module (Automatic Speech Recognition) Locally deploys ASR models to convert speech to text for subsequent processing.

  • LLM Module (Large Language Model) Deploys the Deepseek R1 model locally to process and understand text input, then generate responses.

  • TTS Module (Text-to-Speech) Runs a local speech synthesis model to convert text responses from the LLM into spoken output.

  • Audio Playback Module Outputs audio through an external speaker to complete the interaction loop.

Environment Setup​

Step 1: Download the Code​

asr-llm-tts.zip

Extract the files:

unzip asr-llm-tts.zip -d ~/

Step 2: Install System Dependencies​

Update the system and install required packages:

sudo apt update
sudo apt install -y python3-spacemit-ort spacemit-ollama-toolkit virtualenv wget

Step 3: Install Python Dependencies​

  1. Create a Python Virtual Environment:

    virtualenv ~/demo-env

  2. Configure pip to Use Spacemit Mirror Source:

    pip config set global.extra-index-url https://git.spacemit.com/api/v4/projects/33/packages/pypi/simple

  3. Install project dependencies

    cd ~/asr-llm-tts
    source ~/demo-env/bin/activate
    pip install -r requirements.txt

Step 4: Deploy the LLM Model to Ollama​

  1. Ensure Ollama Service is Running:

    • Verify the service status

      systemctl status ollama

    • Expected output: image-20250422140656034

    • If the status is inactive, start the service:

      systemctl start ollama

  2. Download the Deepseek model and configuration files

    mkdir -p ~/my-ollama
    cd ~/my-ollama
    wget https://archive.spacemit.com/spacemit-ai/openwebui/deepseek-r1-1.5b.modelfile
    wget https://www.modelscope.cn/models/ggml-org/DeepSeek-R1-Distill-Qwen-1.5B-Q4_0-GGUF/resolve/master/deepseek-r1-distill-qwen-1.5b-q4_0.gguf

  3. Load the model into Ollama

    ollama create deepseek-r1-1.5b -f deepseek-r1-1.5b.modelfile

  4. Copy NLTK data

    cp -r ~/asr-llm-tts/nltk_data ~/

Launch Commands​

  • Navigate to the project directory and activate the virtual environment

    cd ~/asr-llm-tts/src
    source ~/demo-env/bin/activate

  • Launch the voice output version

    python main_tts.py

  • Launch the text output version

    python main.py