Aizip has collaborated with Bosch Sensortec to develop the ZenVoice Bone, a cutting-edge deep noise reduction algorithm designed to enhance voice clarity in true wireless stereo (TWS) earbuds, even in extremely noisy environments. Leveraging Bosch Sensortec's BMA550 sensor capabilities, ZenVoice Bone excels in noise reduction, interference separation, and voice capture, while operating efficiently on low-power devices.
ZenVoice Bone is an innovative deep noise reduction algorithm designed to operate on processors costing less than $1, delivering unparalleled voice clarity even in highly noisy settings (above 85db). This technology leverages Bosch Sensortec's high bandwidth bone conduction sensor BMA550 alongside a conventional microphone to accurately capture human speech while filtering out background noise.
Aizip's Zenet neural network architecture complements the BMA550 by providing highly efficient audio processing capabilities. ZenVoice Bone models achieve a remarkable 10-fold reduction in both processing power and memory requirements without compromising noise-handling efficiency. This integration allows the deployment of sophisticated audio intelligence on compact IC chips, significantly reducing costs and power consumption.
According to Aizip, the ZenVoice Bone algorithm maintains underlying speech while removing most noise, even in signal-to-noise ratios as low as -15 dB. The solution can distinguish an end user's voice from background speech without requiring voice registration, a feat impossible with single microphone solutions. And ZenVoice Bone operates efficiently on resource-restricted hardware, with models ranging from 60KB to 300KB in memory footprint, supporting real-time noise elimination with minimal latency, tailored for endpoint and edge device applications.
The BMA550's low noise and wide bandwidth capture a broad spectrum of voice information, essential for the ZenVoice Bone's ability to restore speech amidst loud background noises, such as wind at 85db. The solution's modest computational demand makes it suitable for low-power, compact devices like TWS earbuds and hearing aids.
ZenVoice Bone's ability to deliver clear communication in noisy settings positions it as a transformative technology for various hearable devices, including TWS earbuds, hearing aids, and headsets. The integration of Bosch Sensortec's BMA550 sensor and Aizip's software not only enhances user experience but also broadens the potential applications of hearable devices by extending battery life and reducing cost.
For example, the spectrogram below demonstrates the ZenVoice Bone's effectiveness in a high-noise environment, where it isolates and reconstructs an end user’s voice amidst 85db wind noise.
“The BMA550 sensor data, featuring ultra-low noise and very wide bandwidth for bone conduction, significantly enhances the ZenVoice Bone algorithm's ability to recover important voice information from background noise, ensuring clear voice output,”says Yuan Lu, Co-founder and President at Aizip.
Conversely, the advanced neural network architecture developed by Silicon Valley-based, Aizip brings out the full potential of the BMA550 sensor, showcasing how integrated solutions can drive technological progress. This collaboration highlights the importance of integrating cutting-edge sensor technology with advanced AI solutions to meet evolving consumer demands and set new standards in the audio technology industry.
"Collaborations with experts like Aizip, who specialize in neural networks for hearable applications, enable Bosch Sensortec to deliver sophisticated solutions. By integrating our partners' domain expertise with our advanced sensor technology, we create innovative offerings for our customers," says Marcellino Gemelli, Head of Global Business Development at Bosch Sensortec.
www.aizip.ai
www.bosch-sensortec.com
