Abstract: Detecting a snore seems like easy work, it’s a fairly distinct sound in an otherwise (more or less) quiet environment. Building on (a Wikipedia level understanding of) the work of Pavlov, I seek a method in which to “train” myself to stop snoring- primarily assisted by a laptop, a bluetooth controlled dog shock collar, and some data science. It turns out, when one is about to start electrocuting oneself for snoring the entire process gets very murky, false positive detection errors are “unwanted”, researcher/subject biases are abound, etc. This talk is the journal of the explorations of a total novice audio analyst, seeking to correctly identify snores, and shock himself appropriately.
In this session, attendees will learn about challenges practitioners face at low level implementations of NLP algorithms, strengths and weaknesses of various approaches, and overall how to detect snores. Additionally, attendees will learn about creating full AI systems which operate and learn in an environment with which they interact. Finally, attendees will be entertained throughout the session with various short videos of the speaker electrocuting himself when various attempts failed or succeeded.
While the difficulty level is listed as “intermediate”, it might be better to think of this as a wide array of beginner topics such as (1) The scientific method, and how to approach problems (2) Audio Analysis with Python (3) Full Stack Engineering (and issues that arise such with ‘Full Stack’ thinking (4) Introduction to support vector classification, neural networks / deep learning (LSTMs) and other methods used (5) Hacking and controlling bluetooth devices with Python, and more.
Bio: Trevor Grant is a PMC Member of the Apache Mahout and Apache Streams projects. In his day job he is an Open Source Evangelist / AI Engineer at IBM. He has various videos, blogs, rants, advanced degrees, and code which can all be easily found online against his common handle “rawkintrevo”.