🎤 Social Sidekick

Social Sidekick is a real-time voice-to-AI assistant that listens to your speech, transcribes it using Whisper, and responds with helpful, witty responses generated by the LLaMA 3 model running locally with Ollama. It’s ideal for social practice, conversational exploration, or just having a clever sidekick by your side.

---

✨ Features

• 🗣️ Record your voice using a microphone
• 🎧 Press Enter to start and stop listening
• 📝 Transcribe speech to text using Faster Whisper
• 🤖 Generate smart, funny, or friendly responses using llama3 via Ollama
• 🪵 Real-time console logging to guide your interactions

---

🖥️ Requirements

• Windows, macOS, or Linux
• Python 3.8 or later
• A working microphone
• Ollama installed and running
• Basic terminal usage

---

🛠 Installation Instructions

1. Clone the Project

git clone https://github.com/yourusername/social-sidekick.git
cd social-sidekick

2. Set Up Python Virtual Environment (Recommended)

python -m venv venv

• Windows:

  venv\Scripts\activate

• macOS/Linux:

  source venv/bin/activate

3. Install Required Python Packages

pip install sounddevice scipy numpy keyboard faster-whisper ollama

If you run into permission errors, try pip install --upgrade pip first.

4. Install and Run Ollama

• Download Ollama from ollama.com/download
• Pull the LLaMA 3 model:

ollama pull llama3

• Start the model server:

ollama run llama3

Keep this terminal open in the background while running the Python script.

---

🚀 How to Use

Run the Application

python SocialSidekick.py

Interaction Flow

1. You'll see:

   🔁 Loading Whisper model...
   🎙️ Press [Enter] to start listening.
   

2. Press Enter → Speak → Press Enter again to stop.

3. The app will:

   • Transcribe what you said
   • Send it to LLaMA 3 for a response
   • Show the result like this:

   👂 Heard: What do I say when someone invites me to a group dinner?
   💬 Sidekick: Just say "Thanks for the invite! I'd love to join. What's the plan?"
   

4. It waits for you to press Enter again to repeat the cycle.

---

📷 Example Output

🔁 Loading Whisper model...
🎙️ Press [Enter] to start listening.
🎧 Listening... Press [Enter] again to stop.
🛑 Stopped listening.

👂 Heard: I'm not sure how to say hi to people I don't know.
🤖 Generating reply...

💬 Sidekick: Start with a warm “Hey there! I’m [Your Name], nice to meet you!” It’s classic, friendly, and always works.

---

🧩 Customization

You can adjust the following:

• Whisper model type ("base", "small", "medium", "large")

  whisper = WhisperModel("base", compute_type="int8", device="cpu")

• Sidekick personality Edit the "system" message in chat_with_llama():

  "You're a helpful and witty social sidekick. Help the user fit in..."

---

🛠 Troubleshooting

❗ ModuleNotFoundError: No module named 'ollama'

Install it:

pip install ollama

❗ OSError: No Default Input Device Available

Make sure:

• Your microphone is connected and not being used by other apps.
• You’re not running in a headless environment (like SSH without audio).

❗ Ollama model not found

Pull the model manually:

ollama pull llama3

---

🧑‍💻 Author

Made with 💬 by [Your Name] — inspired by the challenge of being social in a noisy world.

---

📜 License

This project is licensed under the MIT License. Use freely, share widely.

---

🙋 FAQ

Q: Can I use a different model like Mistral or Gemma with Ollama?
A: Yes! Just change the model='llama3' line to another supported model name in chat_with_llama().

Q: Does this run offline?
A: Yes! Both Whisper (with CPU inference) and Ollama run locally. No cloud API keys needed.