🎶 Local MusicGen Studio

A completely free, local AI music generator powered by Meta's MusicGen and Streamlit.
No paid APIs. No cloud dependency. Just open-source creativity on your own machine.

---

Sample Output

Listen to sample

🌟 Features

• 🧠 Text-to-Music Generation using facebook/musicgen-small

• 🎧 Outputs playable MP3 and WAV files

• ⚙️ Runs fully offline after first model download

• 🤖 Optional local LLM (Ollama + Mistral) to refine prompts

• 💾 Download button for easy export

• 💬 Simple and clean Streamlit UI

---

🧩 Project Structure

`local_musicgen_studio/
├── app.py
├── requirements.txt
├── .env.example
├── README.md
└── outputs/`

---

⚙️ Installation & Setup

1️⃣ Create and activate a virtual environment

`python -m venv venv
venv\Scripts\activate          # (Windows)
source venv/bin/activate       # (Mac/Linux)`

2️⃣ Install dependencies

`pip install -r requirements.txt`

If you see warnings about FFmpeg, follow the setup instructions below.

---

🎵 FFmpeg Setup (for MP3 conversion)

🪟 Windows

`pip install ffmpeg-downloader
python -m ffmpeg_downloader -o "C:\ffmpeg"`

Then add `C:\ffmpeg\bin` to your **System PATH**, and verify:

`ffmpeg -version`

🍎 macOS

brew install ffmpeg

🐧 Linux

sudo apt install ffmpeg

---

🗝️ .env Example

# Optional local LLM for prompt refinement OLLAMA_MODEL=mistral

---

🚀 Running the App

streamlit run app.py

Once launched, open the local URL displayed in your terminal.

---

💡 Example Prompts

• "A calm ambient piano piece with soft reverb and slow tempo"

• "Energetic EDM track with strong kick, synth bass, and fast rhythm"

• "Epic orchestral build with strings, drums, and brass"

• "Chill lo-fi beat with vinyl crackle and warm piano loops"

• "Retro 80s synthwave with analog pads and gated drums"

---

⚙️ How It Works

Step	Component	Description
1	Streamlit UI	Collects prompt and duration
2	(Optional) Local LLM	Refines user's natural prompt
3	MusicGen Model	Generates raw audio tensors
4	Pydub + SoundFile	Converts to WAV/MP3
5	Streamlit	Plays and allows download

---

🧠 Key Libraries Used

Library	Purpose
torch	Core ML backend
transformers	Access to Meta's MusicGen model
soundfile	Writing NumPy arrays to WAV
pydub	Convert WAV → MP3
streamlit	Interactive UI
dotenv	Environment variable loader

---

🧰 requirements.txt

`torch==2.2.2
transformers==4.42.3
accelerate==0.31.0
soundfile==0.12.1
pydub==0.25.1
streamlit==1.37.0
python-dotenv`

---

🛠 Customization Tips

• Change facebook/musicgen-small → facebook/musicgen-medium for higher-quality generation.

• Adjust duration_sec slider range for longer or shorter tracks.

• Use Ollama with Mistral or Phi for smarter prompt refinement.

• Save generated outputs automatically to the outputs/ directory.

---

🪩 Example Outputs

Prompt	Duration	Result
"Chill lo-fi beat with soft drums and electric piano"	15 sec	Lo-fi instrumental loop
"Epic cinematic build with strings and brass"	20 sec	Orchestral score intro

---

🧡 Acknowledgements

• Meta AI -- MusicGen

• Streamlit

• Hugging Face Transformers

---

📜 License

MIT License --- free for personal and commercial use.
Please credit the open-source authors when sharing.

---

👨‍💻 Author

Mohamed Azarudeen Allabaksh
AI & Data Science | Stony Brook University
📂 GitHub: github.com/azarudeenallabaksh
🔗 LinkedIn: linkedin.com/in/mohamed-azarudeen