Song Recognizer

This project provides a minimal example of training a neural network to recognise short audio clips. It is built on top of PyTorch and torchaudio and exposes a small command line interface for training and prediction.

Features

• Convert audio files to Mel-spectrograms on the fly
• Simple convolutional neural network architecture
• CLI commands for training and live microphone prediction

Installation

1. Create a virtual environment (optional but recommended)
2. Install the dependencies

pip install -r requirements.txt

Usage

Training

Place your audio files (e.g. WAV or MP3) in a directory and run:

python main.py train /path/to/audio

The trained model will be saved to song_recognizer.pth.

Prediction

To make a prediction using the microphone run:

python main.py predict

or provide a prerecorded file:

python main.py predict --input_file sample.wav

Project Structure

.
├── song_recognizer
│   ├── __init__.py
│   ├── data.py
│   ├── model.py
│   ├── recognition.py
│   └── train.py
├── main.py
├── requirements.txt
└── README.md

License

MIT