Provides a public semantic library for deep customization of industry-specific voice packages for specific industries.
Special industries can deploy their own private cloud services, exclusive access to all voice data, can be used in pure LAN.
The hard noise such as echo and current sound is processed again to eliminate the interference noise; the unique soft noise processing module is very well-recognized with words and homophones to restore the true semantics.
A large public synonym lexicon with no differences in word use and support for private synonym lexicon.
Open companion semantic API + Sanbot robot-side SDK, low user development threshold, no need to develop from scratch, reduce development costs, and provide complete developer technical support.
Parallel access to multiple semantic solutions such as Xunfei, Baidu, IBM Watson, Nuance, etc., users choose semantic results according to their own preferences.
Based on deep learning mode, rely on robot industrial high quality training data accumulation in many years, through millions level sample training, sanbot robot can support multi posture facial detection. Under the extreme light such as dark light and backlight, it still ensures super high stability. This technology is close to robot industrial actual business and can be adapted to wide angle of side face,occlusion, blur, expression changes and other actual environment.Achieving face similarity measurement, it can be used for face clustering, sensitive person monitoring and other scenarios.
Based on the massive face training set collected by Sanbot Cloud Platform, combined with high-dimensional LBP, PCA, LDA and Bayesian, metric learning, migration learning, deep neural network, to achieve better and faster comparison effect.
Achieve the industry's largest thousand-layer face recognition deep learning network. Increase the speed of technology iterations by a factor of five with RapidFlow, the industry's leading deep learning platform. Can achieve billion-level identity face retrieval, millisecond retrieval speed, quasi-real-time return results.