sensiBel and Aizip Deliver Superior Physical AI to Speech Applications, Improving Clarity and Intelligibility

sensiBel, the leader in optical MEMS sensing, and Aizip, the pioneer in artificial intelligence (AI) applications for endpoint and edge devices, today announced a technology collaboration that combines sensiBel’s industry-leading optical MEMS microphone with Aizip’s edge AI beamforming solution — to deliver dramatic improvements in speech clarity and recognition, even in the noisiest environments.

“The best way to improve AI audio capture at the application level is to use the highest-performance microphone at the input level. We’ve proven this through our work with Aizip, a true industry leader in efficient, robust AI models,” said Kieran Harney, CEO, sensiBel. “The combination of Aizip’s low-power, low-latency audio beamforming and noise-cancellation solution with sensiBel’s optical MEMS microphones provides the high-fidelity acoustic sensing layer for physical AI. Together, we deliver exceptional signal-to-noise ratio and dynamic range for improved speech intelligibility and clarity — giving designers of speech-first applications the best of both worlds.”

“Aizip’s AI beamforming and denoising already allow our customers to isolate a specific person’s speech and suppress unwanted sound,” said Yuan Lu, co-founder and president, AiZip. “sensiBel improved our overall edge AI audio solution that much more. Their optical MEMS microphones deliver the exceptional dynamic range that we want from a microphone partner. And their microphone array development kits were easy to use, allowing us to get up and running with their microphones in record time.”

Aizip integrates AI noise reduction, dereverberation, and directional speech enhancement, significantly increasing the speech quality delivered to automatic speech recognition (ASR) engines. This is especially important in noisy and far-field environments.

Test results validate this improvement, showing >30% reduction in character error rate (CER) when using Aizip. This translates to more accurate transcription and downstream AI understanding in Physical AI applications.

Beyond delivering high-quality enhanced audio, Aizip’s AI — paired with high-performance acoustic sensors from sensiBel — can identify the speaker’s direction and position, as well as near-field/far-field acoustic scenarios. These features provide richer audio context to cloud AI/large language models (LLMs) — facilitating more accurate and natural human-AI interaction.

sensiBel’s family of optical MEMS sensors features a traditional ASIC and MEMS sensor, combined with a precision optical measurement module that captures the displacement of the MEMS-based diaphragm. As the diaphragm moves, the optical path changes, and the microphone converts these changes directly into a digital audio signal.

Due to this unique architecture, sensiBel’s optical MEMS sensors — including its MEMS microphone, which achieves 80dB signal-to-noise ratio (SNR), 146dB SPL acoustic overload point (AOP), and a dynamic range of 132dB — far eclipse the performance of capacitive MEMS sensors.

sensiBel offers end-to-end hardware solutions that support rapid design to deployment. For example, its microphone-array development kits, AURORA™ and POLARIS™, work with Windows, MacOS, Android and iOS devices. They connect to computing hardware via USB-C, allowing 24-bit recording to begin within minutes.