Sax | No Sax – Instrument Detection with PyTorch

New Location:

Starting Feb.2018, PyData Chicago event will be hold at our new sponsor - arity. Please note due to building security checkin, you must sign up this meetup to attend. The building may check your ID for entry. Thanks for your understanding.

Don't forget the new address:
222 W Merchandise Mart Plaza
Suite 875
Chicago, IL 60654

Abstract:
In order to identify an instrument, one must consider the unique qualities of its tone. Does the harmonic activity create a dark or bright timbre? Is the envelope pre-determined, as in the decay of a guitar string, or can it be controlled by the player, like the crescendo of a trumpet? All these qualities can be encapsulated and visualized in a spectrogram of an audio signal. This talk will walk through the process of building a binary classifier model to detect the presence of saxophone in an audio sample by passing spectrograms to a convolutional neural network built with PyTorch, one of the most recent and powerful open-source deep learning libraries available for Python. It will discuss a number of aspects of the PyTorch workflow: creating a dataset class, using the data loader, building a network class, configuring the network layers to work with an input image of specific dimensions, viewing parameters, altering initialization settings, and selecting an optimizer, and cross-validation. It will also touch on broader aspects of the process, such as creating a labeled dataset from scratch, using Librosa to load audio and create spectrograms, and tying samples and labels together using MongoDB.

Bio:
Having spent the majority of his life as a professional musician, David Luther is transitioning into the world of data science. He graduated with a BFA in Jazz Studies from the University of Michigan in 2002, and spent the next 15 years touring with various artists — including tenures with Meat Loaf and Dweezil Zappa — in addition to performing his own music around New York City. The time came for a career pivot; a newfound interest in both programming and machine learning collided with his longstanding love of math and physics, and data science was the clear pursuit. He recently relocated to Chicago, and completed the Metis Data Science bootcamp in December 2017.

Schedule:
6:00 p.m - 6:30 p.m is time for social. Seminar will start at 6:30 p.m.

Our Sponsor: arity ( https://www.arity.com/ )