Xiaomi has made significant strides in the field of sound recognition with its self-developed algorithm, marking a major milestone. Using the AudioSet-2M dataset for training, Xiaomi’s audio tagging model achieved a score of over 50 mAP, positioning it as a leader on the global stage.

The AudioSet dataset, divided into three parts by Google, served as the basis for training Xiaomi’s algorithm. Xiaomi’s model surpassed the 50 mAP threshold within the training dataset, setting a new standard in audio tagging technology. Additionally, Xiaomi has introduced a Mini version of the model, specially designed for resource-constrained scenarios, which performs better than similar models from other organizations.

This breakthrough in sound recognition technology has practical applications for Xiaomi’s smart devices, enhancing the overall user experience. The algorithm excels at identifying various environmental sounds, including baby cries, animal noises, and car engines. It can then represent these sounds in different forms, such as text.

Xiaomi’s robots also benefit greatly from this advanced algorithm. The humanoid robot, CyberOne, can recognize 85 types of environmental sounds and perceive a wide range of human emotions through auditory sensing. The second-generation biomimetic quadruped robot, CyberDog 2, further enhances its dynamic response capabilities with the ability to identify 38 types of environmental sounds.

These advancements in sound recognition technology position Xiaomi as a pioneer in the field, and the applications for this technology are wide-ranging. With its continued dedication to innovation, Xiaomi is shaping the future of audio tagging and setting new standards for the industry.

Sources:
– Xiaomi achieves milestone with self-developed sound recognition algorithm
– Xiaomi’s sound recognition algorithm surpasses 50 mAP threshold for the first time