Music-Analyzer

Music-Analyzer is a Python-based machine learning project for analyzing audio files and predicting their genres. It supports both feature-based ML models (Sklearn) and image-based CNN models using mel-spectrograms. The project also includes a Streamlit Web App for easy demo and exploration.

This project is an upgrade and inspired by my first ML project: music_recommender

---

Sneak Peek

Check out a quick preview of the project:

Song	Mel-Spectrogram Preview

---

Table of Contents

• Features
• Project Structure
• Installation
• Dataset Preparation
• Training Models
• Using the Streamlit App
• App Demo
• File Organization & Git
• Contributing
• License

---

Features

• Extracts audio features (MFCCs, Chroma, etc.) for ML models.
• Generates mel-spectrogram images for CNN-based classification.
• Supports multiple model types:
  • Sklearn model (sklearn_model.pkl) for classical ML.
  • CNN model (cnn_model.pth) for deep learning on spectrograms.
• Streamlit web app for uploading audio files and predicting genres.
• Clean project structure for reproducibility and experimentation.

---

Project Structure

Music-Analyzer/
│
├─ data/
│  ├─ raw/             # Raw audio files (ignored by Git)
│  └─ processed/       # Processed features & spectrograms
│
├─ models/
│  ├─ sklearn\_model.pkl
│  └─ cnn\_model.pth
│
├─ src/
│  ├─ dataset\_builder.py   # Build features dataset and save CSV
│  ├─ train\_sklearn.py     # Train Sklearn ML model
│  ├─ train\_cnn.py         # Train CNN on mel-spectrograms
│  ├─ feature\_extraction.py# Extract audio features & save spectrograms
│  └─ app.py               # Streamlit demo app
│
├─ docs/
│  └─ images/              # Screenshots for README demo
│
├─ .gitignore
├─ requirements.txt
└─ README.md

---

Installation

1. Clone the repository:

git clone https://github.com/Longman-max/Music-Analyzer.git
cd Music-Analyzer

2. Create a virtual environment (recommended):

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

3. Install dependencies:

pip install -r requirements.txt

4. Additional libraries (if not included in requirements.txt):

pip install librosa numpy pandas scikit-learn torch torchvision matplotlib pillow streamlit

---

Dataset Preparation

1. Place your raw audio files in data/raw/.

   • If you have genre folders, you can keep them as subfolders.
   • Example: data/raw/rock/song1.wav, data/raw/jazz/song2.mp3

2. Build the dataset:

python src/dataset_builder.py

• This extracts features for ML and generates mel-spectrograms for CNN.
• Features CSV is saved in data/processed/features.csv.
• Mel-spectrogram images are saved in data/processed/melspecs/.

---

Training Models

Sklearn (Classical ML)

python src/train_sklearn.py

• Trains a Logistic Regression classifier on extracted features.
• Saves the model, scaler, and label encoder in models/sklearn_model.pkl.

CNN (Deep Learning)

python src/train_cnn.py

• Trains a ResNet18 CNN on mel-spectrogram images.
• Saves model weights and class labels in models/cnn_model.pth.

---

Using the Streamlit App

Run the app:

streamlit run src/app.py

Features:

• Upload a .wav or .mp3 file.
• Choose between Sklearn or CNN models.
• View predicted genre and feature/spectrogram visualization.

Layout highlights:

• Clean, modern interface with sidebar model selection.
• Columns display predicted genre and spectrogram side by side for CNN.
• Feature vector expandable view for Sklearn predictions.

---

App Demo

Here’s a quick walkthrough of the app in action:

1. Select a Model

Choose between Sklearn (feature-based) or CNN (spectrogram-based) from the sidebar.

---

2. Upload an Audio File

Upload a .wav or .mp3 file to analyze.

---

3. View Predictions

The predicted genre is displayed along with the visualization:

• For Sklearn, you’ll see the extracted feature vector.
• For CNN, you’ll see the mel-spectrogram and predicted label.

---

File Organization & Git

• Ignore raw audio: data/raw/ is ignored via .gitignore.
• Keep processed data: data/processed/ is tracked for reproducibility.
• Ignore large models: models/*.pth, models/*.pkl are ignored to reduce repo size.

Remove already tracked raw files from Git:

git rm -r --cached data/raw/
git commit -m "Stop tracking raw audio files"

---

Contributing

• Fork the repo and create a feature branch.
• Ensure your code is clean and well-documented.
• Submit pull requests for review.

---

License

This project is licensed under the MIT License. See LICENSE for details.