BISH Bash hosted by MathWorks

MathWorks will be hosting the next BISH Bash on Thursday, May 16th! Please, join us for some talks, networking, and bites. See below for the agenda and abstracts of the talks.

We highly encourage in-person attendance if possible, but we will offer a livestream & recording of the talks as well. Check back shortly before the event for details.

Agenda

• 6:00pm: Networking
• 6:30pm: Spatial audio modeling - Francis Tiong, MathWorks
• 7:00pm: Simulating a clarinet - Stephen Thompson, Penn State University
• 7:25pm: Neural Network deployment tuning - Brenda Zhuang, MathWorks
• 7:50pm: Remixing music for hearing aids - Matthew Daly, MathWorks
• 8:15pm: Sound quality evaluation
• 8:40pm: More Networking

Abstracts

• Spatial audio modeling

MathWorks is renowned for offering valuable resources in signal processing. During this session, I'll showcase a series of examples and functions related to spatial audio modeling and time-frequency analysis. Should you question whether MATLAB possesses a function for a specific task, feel free to ask me, and if it's not already available, I can arrange for its implementation.

• Simulating a clarinet

I will present a physical model that replicates a system similar to a clarinet, providing an accurate physical simulation of the playing characteristics found in reed woodwind instruments. Utilizing Simscape/Simulink from MathWorks, this model simulates the acoustic behavior of the air column that resonates within a woodwind instrument. It is based on the assumption that the blowing pressure from the player's mouth is generated by an acoustic pressure source whose amplitude changes over time. The difference in acoustic pressure across the reed triggers its movement. This, in turn, causes the gap between the reed and the mouthpiece tip to fluctuate, controlling the air flow into the mouthpiece based on the gap's instantaneous size.

• Practical workflow to compress neural networks for deployment

Implementing a neural network on a fixed-point processor introduces the challenge of striking a balance between the processing workload and the accuracy of the outcomes. In this session, we will outline how to efficiently perform fixed-point quantization and pruning using MATLAB's tools. The practical approach is demonstrated through an application of acoustic example. We will also introduce a Python co-simulation framework to boost workflow flexibility and efficiency.

• Remixing music for hearing aids

This paper introduces our system submission for the Cadenza ICASSP 2024 Grand Challenge, which presents the problem of remixing and enhancing music for hearing aid users. Our system placed first in the challenge, achieving the best average Hearing-Aid Audio Quality Index (HAAQI) score on the evaluation data set. We describe the system, which uses an ensemble of deep learning music source separators that are fine tuned on the challenge data.

• Sound quality evaluation

How satisfactory is the sound quality, or conversely, how irritating is the noise? Both subjective and objective assessments can aid in quantifying the experience. For instance, in developing noise reduction strategies, having an instant score would be beneficial for efficient regression tuning. This session will explore various valuable metrics relevant to this field.

Directions
The event will be held on the walk-in level of the building. If you are coming from 101 freeway, take the Great America Pkwy exit. There is a huge parking structure beside the building and it is free. A simple map is attached below.