Abstract: A method includes receiving a plurality of unlabeled audio samples corresponding to spoken utterances not paired with corresponding transcriptions. At a target branch of a contrastive Siamese network, the method also includes generating a sequence of encoder outputs for the plurality of unlabeled audio samples and modifying time characteristics of the encoder outputs to generate a sequence of target branch outputs. At an augmentation branch of a contrastive Siamese network, the method also includes performing augmentation on the unlabeled audio samples, generating a sequence of augmented encoder outputs for the augmented unlabeled audio samples, and generating predictions of the sequence of target branch outputs generated at the target branch. The method also includes determining an unsupervised loss term based on target branch outputs and predictions of the sequence of target branch outputs. The method also includes updating parameters of the audio encoder based on the unsupervised loss term.

Abstract: Disclosed herein is a system for controlling a solid state transformer (SST), the SST comprising an AC-to-DC stage, a DC-to-AC stage, and a DC-to-DC stage coupled between the AC-to-DC stage and the DC-to-AC stage, the DC-to-DC stage comprising one or more DC-to-DC converters. The system comprises a stored energy controller coupled to the AC-to-DC stage, the energy controller configured to control the total amount of stored energy within the capacitors of the SST; a power flow controller coupled to the DC-to-AC stage, the power flow controller configured to control power flow in the SST; and one or more energy balancing controllers each coupled to a corresponding DC-to-DC converter, each energy balancing controller configured to balance energy in the corresponding DC-to-DC converter. In some embodiments, the stored energy controller, the power flow controller and the one or more energy balancing controllers are decoupled from one another.

Abstract: A transformer-transducer model for unifying streaming and non-streaming speech recognition includes an audio encoder, a label encoder, and a joint network. The audio encoder receives a sequence of acoustic frames, and generates, at each of a plurality of time steps, a higher order feature representation for a corresponding acoustic frame. The label encoder receives a sequence of non-blank symbols output by a final softmax layer, and generates, at each of the plurality of time steps, a dense representation. The joint network receives the higher order feature representation and the dense representation at each of the plurality of time steps, and generates a probability distribution over possible speech recognition hypothesis. The audio encoder of the model further includes a neural network having an initial stack of transformer layers trained with zero look ahead audio context, and a final stack of transformer layers trained with a variable look ahead audio context.

Abstract: A transformer-transducer model for unifying streaming and non-streaming speech recognition includes an audio encoder, a label encoder, and a joint network. The audio encoder receives a sequence of acoustic frames, and generates, at each of a plurality of time steps, a higher order feature representation for a corresponding acoustic frame. The label encoder receives a sequence of non-blank symbols output by a final softmax layer, and generates, at each of the plurality of time steps, a dense representation. The joint network receives the higher order feature representation and the dense representation at each of the plurality of time steps, and generates a probability distribution over possible speech recognition hypothesis. The audio encoder of the model further includes a neural network having an initial stack of transformer layers trained with zero look ahead audio context, and a final stack of transformer layers trained with a variable look ahead audio context.

Abstract: A method includes receiving a plurality of unlabeled audio samples corresponding to spoken utterances not paired with corresponding transcriptions. At a target branch of a contrastive Siamese network, the method also includes generating a sequence of encoder outputs for the plurality of unlabeled audio samples and modifying time characteristics of the encoder outputs to generate a sequence of target branch outputs. At an augmentation branch of a contrastive Siamese network, the method also includes performing augmentation on the unlabeled audio samples, generating a sequence of augmented encoder outputs for the augmented unlabeled audio samples, and generating predictions of the sequence of target branch outputs generated at the target branch. The method also includes determining an unsupervised loss term based on target branch outputs and predictions of the sequence of target branch outputs. The method also includes updating parameters of the audio encoder based on the unsupervised loss term.

Abstract: A method includes receiving an input audio signal that corresponds to utterances spoken by multiple speakers. The method also includes processing the input audio to generate a transcription of the utterances and a sequence of speaker turn tokens each indicating a location of a respective speaker turn. The method also includes segmenting the input audio signal into a plurality of speaker segments based on the sequence of speaker tokens. The method also includes extracting a speaker-discriminative embedding from each speaker segment and performing spectral clustering on the speaker-discriminative embeddings to cluster the plurality of speaker segments into k classes. The method also includes assigning a respective speaker label to each speaker segment clustered into the respective class that is different than the respective speaker label assigned to the speaker segments clustered into each other class of the k classes.

Abstract: A method for training a speech recognition model with a loss function includes receiving an audio signal including a first segment corresponding to audio spoken by a first speaker, a second segment corresponding to audio spoken by a second speaker, and an overlapping region where the first segment overlaps the second segment. The overlapping region includes a known start time and a known end time. The method also includes generating a respective masked audio embedding for each of the first and second speakers. The method also includes applying a masking loss after the known end time to the respective masked audio embedding for the first speaker when the first speaker was speaking prior to the known start time, or applying the masking loss prior to the known start time when the first speaker was speaking after the known end time.

Abstract: A method for training a speech recognition model with a loss function includes receiving an audio signal including a first segment corresponding to audio spoken by a first speaker, a second segment corresponding to audio spoken by a second speaker, and an overlapping region where the first segment overlaps the second segment. The overlapping region includes a known start time and a known end time. The method also includes generating a respective masked audio embedding for each of the first and second speakers. The method also includes applying a masking loss after the known end time to the respective masked audio embedding for the first speaker when the first speaker was speaking prior to the known start time, or applying the masking loss prior to the known start time when the first speaker was speaking after the known end time.

Abstract: A streaming speech recognition model includes an audio encoder configured to receive a sequence of acoustic frames and generate a higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The streaming speech recognition model also includes a label encoder configured to receive a sequence of non-blank symbols output by a final softmax layer and generate a dense representation. The streaming speech recognition model also includes a joint network configured to receive the higher order feature representation generated by the audio encoder and the dense representation generated by the label encoder and generate a probability distribution over possible speech recognition hypotheses. Here, the streaming speech recognition model is trained using self-alignment to reduce prediction delay by encouraging an alignment path that is one frame left from a reference forced-alignment frame.

Abstract: A transformer-transducer model for unifying streaming and non-streaming speech recognition includes an audio encoder, a label encoder, and a joint network. The audio encoder receives a sequence of acoustic frames, and generates, at each of a plurality of time steps, a higher order feature representation for a corresponding acoustic frame. The label encoder receives a sequence of non-blank symbols output by a final softmax layer, and generates, at each of the plurality of time steps, a dense representation. The joint network receives the higher order feature representation and the dense representation at each of the plurality of time steps, and generates a probability distribution over possible speech recognition hypothesis. The audio encoder of the model further includes a neural network having an initial stack of transformer layers trained with zero look ahead audio context, and a final stack of transformer layers trained with a variable look ahead audio context.

Abstract: A method for training a speech recognition model with a loss function includes receiving an audio signal including a first segment corresponding to audio spoken by a first speaker, a second segment corresponding to audio spoken by a second speaker, and an overlapping region where the first segment overlaps the second segment. The overlapping region includes a known start time and a known end time. The method also includes generating a respective masked audio embedding for each of the first and second speakers. The method also includes applying a masking loss after the known end time to the respective masked audio embedding for the first speaker when the first speaker was speaking prior to the known start time, or applying the masking loss prior to the known start time when the first speaker was speaking after the known end time.

Abstract: This invention relates to a method of controlling a Solid State Transformer (SST).

Abstract: This invention relates to a method of controlling a Solid State Transformer (SST).

Abstract: There is provided a power converter system including a power bus, a plurality of power converter modules connected to the power bus in parallel, a plurality of energy storage modules, each energy storage module coupled to the power bus via a corresponding one of the plurality of power converter modules, and a controller module configured to control at least one of the power converter modules to operate in one of a plurality of operating modes. In particular, the plurality of operating modes of the power converter module includes a plurality of charging power conversion modes for connecting an input power source to the corresponding energy storage module for charging power to the corresponding energy storage module. There is also provided a corresponding method of manufacturing the power converter system.

Abstract: A method for diagnostic monitoring of a wind turbine generator system, said wind turbine generator system comprising a generator, a drive train, and a number of sensors for providing signals and a control system. The method comprising the selection of three sets of signals from said sensors. From the three sets of signals three conditioned sets of signals are obtained by high pass filtering said first set of signals, low pass filtering the second set of signals, and forming a moving average value on the third set of signals. Based on each of said first, second and third set of conditioned signals an evaluation is performed in order to determine a fault, where said evaluation comprises comparing the first, second and third set of conditioned signals with reference values. If said comparison indicates a fault an alarm is set.

Abstract: A method for controlling a variable speed wind turbine generator is disclosed. The generator is connected to a power converter comprising switches. The generator comprises a stator and a set of terminals connected to the stator and to the switches of the power converter. The method comprises: determining a stator flux reference value corresponding to a generator power of a desired magnitude, determining an estimated stator flux value corresponding to an actual generator power, determining a difference between the determined stator flux reference value and the estimated stator flux value, and operating said switches in correspondence to the determined stator flux reference value and the estimated stator flux value to adapt at least one stator electrical quantity to obtain said desired generator power magnitude.

Abstract: A power dissipating arrangement for dissipating power from a generator in a wind turbine is provided. The generator comprises a plurality of output terminals corresponding to a multi-phase output. The power dissipating arrangement comprises a plurality of dissipating units, a plurality of semiconductor switches, a trigger circuit for switching the semiconductor switches and a control unit for controlling the operation of the trigger circuit, thereby controlling the switching of the semiconductor switches.

Abstract: A variable speed wind turbine is provided. The wind turbine comprises a generator, a power converter for converting at least a portion of electrical power generated by the generator, an energy management arrangement coupled to the power converter, the energy management arrangement comprises an energy storage unit, and a controller. The controller is adapted to detect a power imbalance event and to transfer at least a portion of excess electrical energy generated by the generator to the energy storage unit to be stored therein when the power imbalance event is detected.

Abstract: A method of controlling a wind turbine generator is provided, the wind turbine generator converting mechanical energy to electrical. The method comprises: determining an electromagnetic power reference representing the electromagnetic power generated by the wind turbine generator, wherein the electromagnetic power reference is determined based on a desired output of the wind turbine generator; controlling the electrical power generated by the wind turbine generator using a control signal, wherein the control signal is derived from the electromagnetic power reference and is modified in dependence on an inverse power function of the wind turbine generator by incorporating minimal copper loss constraint and stator voltage limiting constraint such that non-linearity of the wind turbine generator is compensated in the control loop and it operates at its maximum efficiency. One effect of the method is that classical linear control loop design can be employed in spite of the plant being a non-linear identity.

Abstract: Parameters of PM machines, especially for IPM machine, are known to vary by significant amounts. This affects the controllability of such machines, which may lead to reduced power loading capability and increased losses. The present invention relates to a method for PM machine inductance profile identification based on voltage mode stator flux observation which could be easily integrated to the generator start-up process in wind turbine application for both stator flux vector feedback control system and current vector feedback control system.