︎

︎ Info
[2021 - Present]
Speech2Spikes
A real-time pre-processing algorithm that enables speech recognition on neuromorphic processors
(Accenture Labs)
MechSense
A fully multi-material 3D printed sensor for revolute motion
(Accenture Labs)
Fire Probe
An IoT sensor for wildfire detection, mitigation, and risk assessment
(Accenture Labs)

A smart joint sensor for measuring performance and chronic pain
(Work in Progress)
[2020 - 2021]
The Ventilator Project
Affordable ventilators to combat a global pandemic
Brain Switch
Brain-computer interface for ALS patients with real-time machine learning
(MIT Media Lab)
AttentivU
Wireless glasses that measure biopotentials and promote well-being
(MIT Media Lab)
Printed Monitors
3D printed studio monitors with flat, full-range sound and excellent off-axis response

HYDROGEL
Open-source 3D printer for printing fluid materials in hydrogel support
(Work in Progress)
If I Kiss You
Sculptural art from hundred-year old paper player piano scroll

[Pre-2020]
Metamaterial strain gauge for soft robots with high signal-to-noise ratio and orthogonal force rejection
Watchtower Robotics
MIT startup fixing water infrastructure using soft robots
(Techstars, MassChallenge)
CNC Controller
Three axis controller with motion control over Ethernet

About Me {
I'm a computer scientist combining physical sensors with intelligent algorithms to create smart systems that improve people's well-being.

My work involves neuromorphic computing, brain-computer interfaces,
3D printing, & IoT devices.

I'm a researcher in the Future Technologies Group @ Accenture Labs.
}

Mark

[ Speech2Spikes ]


S2S / 1001
Sequential stages of data processing (16 KHz 16-bit raw audio, Sliding Discrete Fourier Transform (SDFT), Log-Mel features, Step-Forward spike encoding)

Neuromorphic processors mimic the neural structure and parallel processing capabilities of the human brain, enabling more efficient computing for pattern recognition, computer vision, and other AI applications. Despite the maturity and availability of speech recognition systems, speech recognition has yet to be widely deployed onto neuromorphic systems. This is due to the sparse, spiking nature of these chips being fundamentally different from the continuous, high-resolution form of raw audio.

To translate between the two, we developed Speech2Spikes, an efficient audio processing pipeline that encodes recorded audio into spikes and is suitable for real-time operation with low-power neuromorphic processors. Speech2Spikes is made up of several sequential transformations, each extracting specific pieces of information from the underlying signal. These transformations are not only quite simple, but are readily accelerated by hardware and are capable of being set up as a filter for sample by sample processing.

Published in NICE ’23 (https://doi.org/10.1145/3584954.3584995)