Guest
tinyML Asia 2021 Video Poster Cyberon DSpotter: A phoneme-based local voice recognition solution Alex LIOU, Vice President , Cyberon Corp With speech recognition capability becoming popular in tiny, resource-constrained edge and endpoint IoT devices, Cyberon proposes an offline algorithm with a small memory footprint and low power consumption, called DSpotter, for always-on voice trigger and multiple command sets recognition. It uses phoneme-based approach, enabling developers to create a speaker-independent (SI) voice recognition model without requiring a costly training data collection process for specific commands to achieve high performance. The adoption of DNN-based modeling also ensures DSpotter has high robustness and accuracy under noise conditions even without speech preprocessing. DSpotter is equipped with more than 40 global language coverage, which facilitates developers to utilize voice technology on edge devices and get to global market in time.
