Acoustic Mirror Product | 2023.11-2024.07
Research commercialization – Compact acoustic detection device using 64-mic array for sound source localization.
- Technical Leadership: Responsible for overall architecture, algorithm, procurement, and structure development; guided hardware and software development.
- Project Management: Broke down the project into algorithm, software, structure, hardware, and IP modules, used OKR planning to define phases and key milestones, and assigned tasks to team members.
- Team Management: Overall project advancement, solving development challenges, and proactively seeking external resources when needed.
Outcome: Prototype developed in 3 months, pending user testing.
DJI-AUDIO Common Algorithm / OSMO Camera Audio Implementation | 2020.04-2023.08
Common algorithm maintenance and product implementation for handheld cameras.
- Integrated complete audio signal processing algorithm with up to 8-channel input and 2-channel output, including bypass, preprocessing, array algorithm, dynamic gain, etc.
- Audio algorithm implementation for DJI Pocket2, ACTION2, ACTION3, and DJI MIC, responsible for sound source localization, directional pickup, zoom pickup, enhanced stereo, dynamic gain, wind noise, mic blockage, time alignment, circuit noise, and other modules.
Small Array Spatial Audio Capture & Playback (Pre-research) | 2021.04-2022.08
Implemented spatial audio capture using 3 mics with diameter under 30mm and applied it to spatial sound collection; achieved binaural playback using Head-Related Transfer Function (HRTF).
Role: Pre-research project lead, responsible for development and progress reporting. Deployed DMA algorithm to compact cameras.
Outcome: Demo completed, spatial audio capture and binaural playback achieved, and tuned audio delivery to subsequent colleagues.
Drone Audio Capture (Pre-research) | 2022.04-2023.08
Implemented hover-state drone audio capture using 8 mics spaced 10mm apart.
- Built an 8-mic array platform and implemented cardioid differential array algorithm.
- Proposed a solution for low-frequency white noise gain issues and applied it to the algorithm model.
- Gain/attenuation performance comparable to Rode VideoMic in end-fire directions.
- Applied the algorithm to drone-mounted arrays, achieving intelligible shout recognition at 10 meters without noise reduction while hovering.
- Also applied to robot vacuum body audio capture with significant improvement.
Outcome: Drone hover audio intelligible; robot vacuum audio SNR improved by 10dB.