Abstract: A method for training a sub-model for contextual biasing for speech recognition includes obtaining a base speech recognition model trained on non-biased data. The method includes obtaining a set of training utterances representative of a particular domain, each training utterance in the set of training utterances including audio data characterizing the training utterances and a ground truth transcription of the training utterance. The method further includes, for each corresponding training utterance in the set of training utterances, determining, using an embedding encoder, a corresponding document embedding from the ground truth transcription of the corresponding training utterance. The method includes training, using the corresponding document embeddings determined from the ground truth transcriptions of the set of training utterances, a sub-model to bias the base speech recognition model to recognize speech in the particular domain.

Abstract: Disclosed embodiments may include a method that includes receiving profile data associated with a user, the profile data comprising a first phone number, receiving browsing data from a user device associated with the user, storing the browsing data with a store time, receiving a phone call from the first phone number at a call time, identifying the user from the first phone number, retrieving the browsing data, determining whether the store time is within a time threshold of the call time and whether the browsing data comprises trigger data. When the browsing data comprises the trigger data and the store time is within the time threshold of the call time, generate a customized audio IVR menu based on the browsing data and present the customized audio IVR menu to the user over the phone call.

Abstract: A heating assembly for a vapour generating device includes: a heating chamber arranged to hold an aerosol generating medium; a heater arranged, in use, to provide heating to the heating chamber; and an ejector arranged to controllably eject, in use, the aerosol generating medium from the heating chamber. A vapour generating device includes the heating assembly; and a vapour passage to carry vapour generated in the heating chamber from the heating chamber to an air outlet, wherein at least a portion of the vapour passage is formed by the ejector.

Abstract: Systems and methods of the present disclosure enable anomaly detection based on data lineage by using at least one processor to receive initial lineage records that record initial changes in data items at each data processor of a lineage stream. The processor(s) generates a historical distribution as a data processor baseline for a particular data processor based on each change in each data item at the particular data processor. For lineage records in a subsequent time period, the processor(s) identifies subsequent changes in data items at the particular data processor and generates a real-time dynamic distribution to represent a current data processor behavior for the particular data processor based on each change for the subsequent lineage records. The processor(s) determines a deviation between the data processor baseline and the current data processor behavior based on the historical distribution and the real-time dynamic distribution to identify a data processor anomaly.

Abstract: Implementations set forth herein relate to an automated assistant that is invoked according to contextual signals—in lieu of requiring a user to explicitly speak an invocation phrase. When a user is in an environment with an assistant-enabled device, contextual data characterizing features of the environment can be processed to determine whether a user intends to invoke the automated assistant. Therefore, when such features are detected by the automated assistant, the automated assistant can bypass requiring an invocation phrase from a user and, instead, be responsive to one or more assistant commands from the user. The automated assistant can operate based on a trained machine learning model that is trained using instances of training data that characterize previous interactions in which one or more users invoked or did not invoke the automated assistant.

Abstract: A closure system including a cap and a spout, where the cap and/or spout contain at least one discrete layer of barrier material. In one example, the spout includes a conduit and a seal member, the conduit being arranged about an imaginary axis and the seal member having one or more surfaces configured to be secured to a container. The cap includes an inner circumferential surface, the inner circumferential surface arranged to encompass the conduit. The cap also includes a first discrete layer of thermoplastic resin and a first discrete layer of barrier material and the seal member includes a second discrete layer of thermoplastic resin and a second discrete layer of barrier material.

Abstract: The present disclosure provides semirigid pouch designs and methods of manufacturing pouches. Such pouches can be made of plastic or non-plastic materials and may be easily sorted and recycled within existing rigid packaging recycling streams. The pouches are semirigid, while flexible, and will substantially retain their shape during and following product dispensing. This is accomplished by sealing a lower edge of a wall of semirigid plastic or non-plastic materials to a rigid or semirigid base, enabling the pouch to stand up on the base. Portions of an upper edge of the wall are affixed together to form a space to hold the foodstuffs. Further, the present pouch design does not require an additional label to be applied by the manufacturer, as the present design provides a large, uninterrupted, surface area for printing of advertisements and/or labels directly to the outer surface of the pouch.

Abstract: Disclosed is a DC power supply device that includes: a transformer having a primary side and a secondary side; a diode rectifier connected at its input side to the secondary side of the transformer; an inverter connected at its output side to the secondary side of the transformer; and a controller. The inverter is controlled by the controller to generate reactive power and/or harmonics onto the secondary side of the transformer so as to regulate the DC voltage at the output side of the diode rectifier to a target value. The controller receives at its input side at least one DC signal outputted by the diode rectifier and uses the at least one DC signal to control the inverter.

Abstract: A portable aerosol-generating device for receiving an aerosol-generating substance includes a chamber arranged to receive an aerosol-generating substance and an aerosol-generating unit configured to generate an aerosol from the aerosol-generating substance for a user to inhale in use. The chamber is configured to be releasably inserted into the device through an opening in the device. This allows the chamber to be removed from the device and cleaned between consecutive uses of the device.

Abstract: A spill-resistant spout and method of injection molding a spill-resistant spout. The spill-resistant spout includes a conduit portion having an exit aperture in fluid communication with a central channel; and a seal portion having a bottom surface, the seal portion secured to the conduit portion and in fluid communication with the central channel, the bottom surface comprising at least two intake apertures configured to receive foodstuffs from a container. The use of multiple intake apertures increases the surface tension on the foodstuffs attempting to pass through the intake apertures while maintaining or increasing the total flow volume of foodstuffs through the spill-resistant spout. Additionally, the barrier qualities of the spill-resistant spout can be enhanced by providing one or more discrete layers of thermoplastic material and one or more layers of barrier material within at least one portion of the spill-resistant spout.

Abstract: Systems, methods, devices, and other techniques are described herein for determining dialog states that correspond to voice inputs and for biasing a language model based on the determined dialog states. In some implementations, a method includes receiving, at a computing system, audio data that indicates a voice input and determining a particular dialog state, from among a plurality of dialog states, which corresponds to the voice input. A set of n-grams can be identified that are associated with the particular dialog state that corresponds to the voice input. In response to identifying the set of n-grams that are associated with the particular dialog state that corresponds to the voice input, a language model can be biased by adjusting probability scores that the language model indicates for n-grams in the set of n-grams. The voice input can be transcribed using the adjusted language model.

Abstract: A spill-resistant spout and method of injection molding a spill-resistant spout. The spill-resistant spout includes a conduit portion having an exit aperture in fluid communication with a central channel; and a seal portion having a bottom surface, the seal portion secured to the conduit portion and in fluid communication with the central channel, the bottom surface comprising at least two intake apertures configured to receive foodstuffs from a container. The use of multiple intake apertures increases the surface tension on the foodstuffs attempting to pass through the intake apertures while maintaining or increasing the total flow volume of foodstuffs through the spill-resistant spout. Additionally, the barrier qualities of the spill-resistant spout can be enhanced by providing one or more discrete layers of thermoplastic material and one or more layers of barrier material within at least one portion of the spill-resistant spout.

Abstract: Disclosed herein are system, method, and computer program product embodiments for performing in-flight transformations of data streams. Complex event processing (CEP) operating on event records from various data streams can operate by storing the streamed event record data in a database, and then querying the data to perform a data transformation. However, in order to improve the performance of the data transformations and streamline CEP, a serverless architecture is introduced that can perform data transformations directly on streams, using attributes of the streamed record data defined in a schema. The resulting transformed data can then be provided by the serverless architecture to the CEP for direct access of the data most needed by the CEP.

Abstract: In certain embodiments, machine learning and lineage data may be used to manage data. In some embodiments, a computing system may use lineage data to identify two datasets that may be related. The computing system may determine that a user has access to a derivative dataset but does not have access to an original dataset that was used to create the derivative dataset. In response, the computing system may use a machine learning model to generate a similarity score indicating a level of similarity between the original dataset and the derivative dataset. If the similarity score satisfies a threshold score, the computing system may modify access rights of the user so that the user is unable to access a portion of the data in the derivative dataset.

Abstract: Systems, methods, devices, and other techniques are described herein for determining dialog states that correspond to voice inputs and for biasing a language model based on the determined dialog states. In some implementations, a method includes receiving, at a computing system, audio data that indicates a voice input and determining a particular dialog state, from among a plurality of dialog states, which corresponds to the voice input. A set of n-grams can be identified that are associated with the particular dialog state that corresponds to the voice input. In response to identifying the set of n-grams that are associated with the particular dialog state that corresponds to the voice input, a language model can be biased by adjusting probability scores that the language model indicates for n-grams in the set of n-grams. The voice input can be transcribed using the adjusted language model.

Abstract: An apparatus for controlling a vapour generating device, which generates vapour through use of a consumable, includes a data storage medium and a communication means for communicating with the device. The data storage medium is arranged to store data associated with control of activation of the device and data associated with control of the vapour generating device in use. The communication means communicates with the device such that, in use, both activation and operation of the device can be controlled based on the stored data.

Abstract: Methods, media and systems for screening content for malicious content in a web browser are disclosed. Document object model (DOM) elements are retrieved from the content displayed in the web browser. A link to visit a web site included in the DOM elements is identified. The identified link is compared to a list of permitted sources. Information is scanned to determine whether the information matches records of content transmitted by a primary source among the permitted sources. An alert is generated after determining a mismatch between the information in the DOM elements and records of content transmitted by the primary source.

Abstract: Disclosed embodiments may include a method that includes receiving profile data associated with a user, the profile data comprising a first phone number, receiving browsing data from a user device associated with the user, storing the browsing data with a store time, receiving a phone call from the first phone number at a call time, identifying the user from the first phone number, retrieving the browsing data, determining whether the store time is within a time threshold of the call time and whether the browsing data comprises trigger data. When the browsing data comprises the trigger data and the store time is within the time threshold of the call time, generate a customized audio IVR menu based on the browsing data and present the customized audio IVR menu to the user over the phone call.

Abstract: A dataset is received for ingestion into a data platform, and a correlation identifier is generated responsive to receiving the dataset. Multiple choreographed services emit multiple event messages. The plurality of choreographed services operate independently of each other based on a plurality of events triggered in a data platform. The plurality of events relate to contents of the dataset and comprising the correlation identifier. A message storage is populated with multiple status updates related to the correlation identifier. A status message associated with the correlation identifier is published in response to a status update of the plurality of status updates.

Abstract: In certain embodiments, machine learning and lineage data may be used to manage data. In some embodiments, a computing system may use lineage data to identify two datasets that may be related. The computing system may determine that a user has access to a derivative dataset but does not have access to an original dataset that was used to create the derivative dataset. In response, the computing system may use a machine learning model to generate a similarity score indicating a level of similarity between the original dataset and the derivative dataset. If the similarity score satisfies a threshold score, the computing system may modify access rights of the user so that the user is unable to access a portion of the data in the derivative dataset.