Recent DSP Projects
Recent DSP Projects
Embedded Files
1. Audio Detection and Direction of Arrival Estimation w/ ESP32 + Digital Mics (In Progress)
1. Audio Detection and Direction of Arrival Estimation w/ ESP32 + Digital Mics (In Progress)
Project Overview: Sound detection subsystem for a wearable safety device that alerts pedestrians to approaching vehicles through real-time audio processing of car horns and sirens.
Major Design Decisions + Features:
FFT-based frequency detection: Converts incoming audio into frequency spectrum to identify car horn patterns (600-1,500 Hz range); compares energy in the horn frequency band against total spectrum energy - when horn frequencies contain 30%+ of total power, triggers alert (filters out wind noise and general traffic)
Real time audio processing: Samples audio at 16 kHz, processes 512-sample chunks through FFT with Hamming window smoothing, entire pipeline from sound capture to wireless alert completes in under 100ms
Field tested: Achieves 90%+ detection rate for vehicle horns within 15 meters in real-world conditions
ESP32 dual-core architecture: Uses one processor core for continuous FFT computation while the second handles ESP-NOW wireless protocol, transmitting alerts to wearable haptic feedback unit (haptics integration currently in progress)
2. Dynamic Video Editing Engine in Julia
2. Dynamic Video Editing Engine in Julia
Project Overview: This project uses Julia-based audio analysis to detect musical structure in a song, including onsets and tempo, and then uses those timing cues to drive when visual edits happen. It essentially synchronizes visuals to the music.
Major Design Decisions + Features:
Built a full STFT pipeline in Julia. Takes input wav file and converts to mono, applies Hann-windowed framing, real FFT with FFTW, and log-compressed magnitude spectrogram to make onsets more pronounced.
Used spectral flux as the core onset function, then stabilized it with EMA smoothing and a moving percentile/median baseline, so the detector adapts to changing loudness over time.
Implemented a custom peak-picking stage with local maxima, thresholding above the baseline, and a configurable refractory period that keeps only the strongest onset per window.
Estimated a global BPM from the flux signal using autocorrelation. BPM plots for debugging and interpretation.
After I finish coding the audio detection features in Julia, my goal is to build a Julia-to-Java pipeline to drive dynamic video editing synchronized with the audio.
Dynamic_Video_Editor_Demo.MOV3. 5-Band Audio Equalizer in MATLAB
3. 5-Band Audio Equalizer in MATLAB
Major Design Decisions + Features:
Designed a simple 5-band graphic EQ in MATLAB using FIR filters built with fir1 (low-pass, band-pass, and high-pass), with band edges chosen to roughly cover bass, low-mids, mids, high-mids, and treble.
Mixed stereo input down to mono to keep the signal path and filter design straightforward.
Applied user-set gains in dB per band (converted to linear scale), filtered each band separately, then summed all bands back together and normalized to avoid clipping.
Added plotting for both time-domain waveforms and magnitude spectra, plus frequency responses of each band’s filter, so I can see how the EQ is shaping the sound.
I ran the script on a sample punk metal audio using gain = [-5, 0, 0, 0, 0], which effectively attenuates the lowest band (0–200 Hz) by 5 dB; all other bands are left flat. It cuts the bass a bit. See plots below:
Page updated
Report abuse