Misaki is a lightweight offline desktop voice assistant built in Python that runs smoothly even on low-end laptops.

Idle	Listening	Thinking	Talking

It listens for a wake word ("Hey Misaki"), understands speech, thinks using a local LLM, and replies with a natural voice — all without internet.

It also includes a cute animated robot face GUI that changes expressions while:

• Idle 😴
• Listening 🎧
• Thinking 🤔
• Speaking 🗣️

✅ Wake word detection
✅ Real-time mic streaming
✅ Silence detection (auto stop recording)
✅ Offline Speech-to-Text ( Whisper )
✅ Local LLM brain ( Ollama – phi3 mini )
✅ Offline Text-to-Speech ( Piper TTS )
✅ Animated GUI with GIF expressions
✅ Works on low-end CPU laptops
✅ No cloud / no API cost
✅ Take Photos and Screenshots (say take picture / screenshot)
✅ Can access internet when ever she want
✅ Mikami also has memory to keep information

MyFriend/
│
├── main.py                   # CLI assistant
├── gui_main.py               # GUI version
│
├── 📂brain/
│   ├── cerebrum.py           # LLM logic (Ollama)
│   ├── 📂stt/
│   │   ├── auditoryCortex.py # Speech → text (Whisper)
│   └── 📂tts/
│       └── motorCortex.py    # Text → speech (Piper)
├── 📂face/                  # GIF animations
│   ├── Idle.gif
│   ├── listening.gif
│   ├── thinking.gif
│   └── Talking1.gif
├── 📂ears/
│   ├── mic.py               # record until silence
│   └── mic_small.py         # wake word chunk recorder
├── 📂ears/
│   ├── gif.py               # GifPlayer
│   └── get_path.py          # resource path helper
├── voice/                   # Piper voices
│
└── README.md

Processor (CPU): 2 GHz or faster.
Memory (RAM): 8 GB or higher.
Storage: 3 GB.

pip install -r requirements.txt

Download: 👉 https://ollama.com

Then:

ollama pull phi3:mini

Download voice files if you want different voice:
👉 https://Piper.com
Example:

en_US-amy-medium.onnx
en_US-amy-medium.onnx.json

Put inside:

voice/

python gui_main.py

python main.py

Say:

Hey Misaki

Then speak your command.

Example:

Hey Misaki
What time is it?

Buddy will:

1. Wake
2. Listen
3. Think
4. Speak reply

If animations lag:

Use:

200x200 px

Flow:

Mic → Whisper (STT)
     → Ollama (Brain)
     → Piper (TTS)
     → Speaker

GUI states:

Idle → Listening → Thinking → Talking → Idle

You can share Buddy with friends.

pip install pyinstaller

pyinstaller --onefile --noconsole gui_main.py

EXE will be in:

dist/gui_main.exe

Buddy runs:

• locally
• no internet
• no data sent anywhere

Safe for personal use.

Aryan Gawade

• 🔗 LinkedIn
• 🔗 GitHub URL
• 🔗 Portfolio