Abstract: Techniques performed by a data processing system for training a Recurrent Neural Network Transducer (RNN-T) herein include encoder pretraining by training a neural network-based token classification model using first token-aligned training data representing a plurality of utterances, where each utterance is associated with a plurality of frames of audio data and tokens representing each utterance are aligned with frame boundaries of the plurality of audio frames; obtaining first cross-entropy (CE) criterion from the token classification model, wherein the CE criterion represent a divergence between expected outputs and reference outputs of the model; pretraining an encoder of an RNN-T based on the first CE criterion; and training the RNN-T with second training data after pretraining the encoder of the RNN-T. These techniques also include whole-network pre-training of the RNN-T.

Abstract: A user equipment (UE) device may perform random access to a base station in the context where propagation delay between the base station and the UE device is large. The UE device randomly selects a random access preamble from an available set of preambles. The UE device selects a time slot for transmission of the random access preamble based on a configured correspondence between allowable time slots and subsets of the available set. The UE device transmits the random access preamble to a base station in the selected time slot, with timing advance to compensate for a common delay. Thus, the base station, knowing the configured correspondence, is able to determine the time slot in which the random access preamble was transmitted. The UE device receives a random access response that has been addressed with a Random Access Radio Network Temporary Identifier (RA-RNTI).

Abstract: Techniques performed by a data processing system for training a Recurrent Neural Network Transducer (RNN-T) herein include encoder pretraining by training a neural network-based token classification model using first token-aligned training data representing a plurality of utterances, where each utterance is associated with a plurality of frames of audio data and tokens representing each utterance are aligned with frame boundaries of the plurality of audio frames; obtaining first cross-entropy (CE) criterion from the token classification model, wherein the CE criterion represent a divergence between expected outputs and reference outputs of the model; pretraining an encoder of an RNN-T based on the first CE criterion; and training the RNN-T with second training data after pretraining the encoder of the RNN-T. These techniques also include whole-network pre-training of the RNN-T.

Abstract: A two-dimensional auto compensated multi-axis tolerance adaptive system used in the cartridge carrier of an inject office machine is constructed to include a carrier unit, a reciprocation mechanism, and a driving mechanism, the carrier unit having a clearance sensor and a bias sensor, the driving mechanism including a rotating unit fastened pivotally with the carrier unit and a lifting unit coupled to the reciprocation mechanism for a linear movement for enabling the plane of rotation of the rotating unit to form with the straight lifting line of the lifting unit a X-Y plane and Z-axis space coordinates system. By means of the detection and scanning of the clearance sensor and the bias sensor, the ink cartridge carrier actively controls the optimum printing clearance between the ink jet nozzle and the media.

Abstract: The present invention provides therapeutically useful substituted guanidines, and methods of treatment and pharmaceutical compositions that utilize or comprise one or more of such guanidines.