Inventor · Beijing, CN

Jinyu Li

100Patents
11h-index
168Co-inventors
79Inventor score

Filing activity: Dec 3, 2007 → Jan 26, 2024

Most-cited inventions

PatentTitleAreaCited byStatus
US9454958B2 Exploiting heterogeneous data in deep neural network-based speech recognition systems Physics 28 Active
US8824749B2 Biometric recognition Physics 27 Active
US9324321B2 Low-footprint adaptation and personalization for a deep neural network Physics 19 Active
US8239195B2 Adapting a compressed model for use in speech recognition Physics 18 Active
US8687880B2 Real time head pose estimation Physics 17 Active
US9251427B1 False face representation identification Physics 17 Active
US9842585B2 Multilingual deep neural network Physics 16 Active
US8711159B2 VGPU: a real time GPU emulator Physics 15 Active
US9703940B2 Managed biometric identity Physics 14 Active
US10839822B2 Multi-channel speech separation Electricity 13 Active
US9728184B2 Restructuring deep neural network acoustic models Physics 12 Active
US8265341B2 Voice-body identity correlation Physics 11 Active
US10347241B1 Speaker-invariant training via adversarial learning Physics 10 Active
US8924735B2 Managed biometric identity Physics 10 Active
US10650226B2 False face representation identification Physics 9 Active
US9539500B2 Biometric recognition Physics 9 Active
US10629193B2 Advancing word-based speech recognition processing Physics 8 Active
US9280969B2 Model training for automatic speech recognition from imperfect transcription data Physics 8 Active
US8864581B2 Visual based identitiy tracking Physics 7 Active
US8217962B2 Single-pass bounding box calculation Physics 6 Active
US9520127B2 Shared hidden layer combination for speech recognition systems Physics 6 Active
US10643602B2 Adversarial teacher-student learning for unsupervised domain adaptation Physics 6 Active
US10885900B2 Domain adaptation in speech recognition via teacher-student learning Physics 5 Active
US11587569B2 Generating and using text-to-speech data for speech recognition models Physics 4 Active
US10019990B2 Variable-component deep neural network for robust speech recognition Physics 4 Active

Source: USPTO / EPO open patent data. Inventor disambiguation is heuristic; counts are objective bibliographic measures.