Objectives

---

For this project you will do the following:
Build dataset containing song metadata, and their various genres, and spectrograph information.
Develop a pipeline to import audio clips from datasets.
Train a CNN or other neural network to predict music genres from a curated dataset.
Develop a program to run a user-submitted audio clip against the model and print results with accuracy metric.

Running the program:
The user will enter a song clip, then receive a formatted top-n list of genres sorted by confidence value in descending order.

The following technologies may be useful:
Keras and Tensorflow to train, test, and validate the neural networks. Librosa for audio pre-processing and manipulation. Matplot for data analysis and visualization. Numpy and Pandas for data structures and utility functions. Youtube-dl to obtain the audio files for the training dataset.

The following datasets may be useful:
A pre-existing dataset for the project include the GTZAN Genre Collection used in the well-known paper in genre classification: “Musical Genre Classification of Audio Signals” by G. Tzanetakis and P. Cook in IEEE Transactions on Audio and Speech Processing 2002.
Also the Million Song Dataset curated by LabROSA and The Echonest.

Stretch goals:
Create a web app front-end (can run on desktop).
Host the program as a web server.
Content based recommender system for music similar to audio clip

Motivations

---

To gain knowledge and experience massaging datasets, processing audio in python, as well as training and evaluating neural networks

Qualifications

---

Minimum Qualifications:
some experience with python3

Preferred Qualifications:
Knowledge of tensorflow and keras

Details

---

Project Partner:

William Pfeil

NDA/IPA:

No Agreement Required

Number Groups:

1

Project Status:

Accepting Applicants

Keywords:
Neural NetworksMusicAudioAIClassificationgenresrecommendertensor flowkeras