I love everything to do with sound.
Welcome to my site.
I love everything involving music, technology, and sound. I work as a machine learning engineer researching and developing new deep learning systems for a range of audio and music technologies. I have also consulted on the acoustical design of the built environment, with projects ranging from performance venues to recording studios. My work and research include developing and implementing models and algorithms to perform audio source separation, music sample generation, sound source detection, zero-shot classification, music fingerprinting, audio enhancement, and binaural audio externalization among others. I have also developed other software to measure acoustics metrics, model wave behavior, process spatial audio, and create audio effects plugins.
When I am not focused on the design of sound and technology through my work or research, I am writing, recording, or performing music. I play guitar (electric, acoustic, bass), percussion, piano, and even dabble with the banjo. My artist name is After August (check me out on Spotify), and I play musical styles from rock to jazz to folk and everything in between. I love all types of music.
I live in Chicago with my wife, Steph, our son, Myles, and our little dog, Queso. Steph works as an interventional radiology nurse at Northwestern Memorial Hospital assisting with groundbreaking operations daily and Queso takes naps on my lap daily as I work at my desk. I play in the local kickball league and love to escape to the Finger Lakes and Adirondack Mountains in NY each summer for some hiking, kayaking, and time with the family.
Leading the end-to-end research and development of state-of-the-art deep learning systems for audio source separation, sound event detection, zero-shot classification, music fingerprinting, and audio enhancement. Drove ML technologies from MVP to industry-leading performance, fueling multi-million-dollar annual recurring revenue growth.
Advanced core product technology by designing and innovating cutting-edge machine learning frameworks in PyTorch, novel model architectures, scalable system designs, and optimized production pipelines—improving performance and inference speed while reducing model sizes by over 90%.
Created and curated extensive proprietary audio datasets. Developed semi-automated labeling tools for high-quality annotations and leveraged advanced data augmentation techniques to synthetically scale data by over 10x, improving model robustness and generalization.
Collaborating on the development and implementation of a real-time binaural externalization algorithm for object-based spatial audio in Max/MSP and RNBO, addressing the unsolved immersive audio problem of improper frontal source elevation perception while minimizing spectral coloration via an all-pass framework.
Developed proprietary software collaboratively with a research team in MATLAB to model wave behavior via the finite-difference time-domain method, producing a new company tool for precise acoustic diffusion analysis.
Built a suite of software tools in MATLAB for impulse response acquisition and general acoustics utilities, increasing company-wide efficiency, accuracy, and capabilities of acoustics measurements and analysis.
Delivered a wide range of successful projects, including performing arts, education, civic, worship, experimental, corporate, residential, and environmental, by managing project teams, timelines, and budgets effectively.
Launched a new product offering by developing a room acoustics auralization system in Max/MSP and MATLAB that can encode, convolve, and decode higher-order ambisonic signals in real-time.
Reduced acoustic design time on projects by updating company protocols to standardized acoustic analysis and testing methods, as well as designing bespoke analysis tools in MATLAB, increasing company-wide efficiency, accuracy, and capabilities of acoustics measurements and analysis.
Researched and patented new technology, by creating a novel machine learning algorithm in MATLAB that estimates the directions of arrival and relative levels of an arbitrary number of sound sources through a multi-level Bayesian framework, using spherical beamforming with a spherical microphone array.
Built a lightweight system that applies waveform diffusion (1D U-Net) to generate short audio samples (such as drum sounds) to be trained and run on a low-level consumer GPU (<2GB VRAM). The purpose of this project is to provide access to waveform diffusion code for those interested in exploration but who have limited hardware resources.
Built a music source separation system in Python using a band-split recurrent neural network (RNN) framework based on a research paper to compete in AIcrowd's Sound Demixing Challenge 2023. Four models (voice, bass, drums, other) were trained on a Google Cloud Platform GPUs with W&B tracking and resulted in an improvement over the baseline model on the "Label Noise" track by 42%.
Developed a PyTorch implementation of logWMSE, an audio quality metric and loss function. It addresses shortcomings of common audio metrics, like digital silence targets, making it effective for training and evaluating audio source separation and denoising models.
Wrote and produced a song using AI-generated audio by leveraging the audio diffusion tool Dance Diffusion. Utilized waveform-based (1D convolution) diffusion with unconditional generation and audio style transfer functions to produce high-quality 48kHz samples.
Read more on my Medium home page.
An examination of the current tensions between AI companies and music creators over training data rights, offering a roadmap toward a more collaborative future built on transparency, consent, and fair compensation.
A deep dive into the principles powering modern generative music AI, demystifying how diffusion models create new sounds and exploring their potential as tools to enhance rather than replace human creativity.
Explains the design decisions behind a compact waveform diffusion model that fits consumer GPUs, covering training tricks, architecture choices, and practical generation workflows.
A look at modern audio source-separation breakthroughs, tracing how open research, competitions, and real-world releases are reshaping what engineers and artists can do with individual stems.
