Abstract: Embodiments are provided for building a configurable multilingual model. A computing system obtains a plurality of language-specific automatic speech recognition modules and a universal automatic speech recognition module trained on a multi-language training dataset comprising training data corresponding to each of the plurality of different languages. The computing system then compiles the universal automatic speech recognition module with the plurality of language-specific automatic speech recognition modules to generate a configurable multilingual model that is configured to selectively and dynamically utilize a sub-set of the plurality of language-specific automatic speech recognition modules with the universal automatic speech recognition module to process audio content in response to user input identifying one or more target languages associated with the audio content.

Abstract: A system obtains a first training data set comprising labeled speech data or both labeled and unlabeled data corresponding to a high-resource data set as well as latent speech representations based on the first training data set. The system trains a machine learning model on the first training data set to learn phonetically aware speech representations corresponding to the first training data set. The system applies the latent speech representations to a transformer context network to generate contextual representations. The system aligns each of the contextual representations with a phoneme label to generate phonetically-aware contextual representations. The system causes a refinement engine to further refine the machine learning model.

Abstract: Aspects of the subject technology provide techniques for improved re-use of digital images, including creation of a new “sticker” image by extracting one or more objects from a prior image. In an implementation, images in a library may be preprocessed to identify image attributes of corresponding images in the library, including suitability scores, classifications of objects in the images, and masks for the classified objects.

Abstract: A device and a method for early monitoring of gas intrusion based on pressure wave propagation are provided in the present disclosure. The technical solution is as follows. The lower end of the liquid storage tank is connected to the simulated wellbore through a liquid injection pipeline, a centrifugal pump, a pressure-stabilizing water tank, and a mass flowmeter. One end of the gas storage tank is connected to a screw air compressor, while the other end is connected to the simulated wellbore through an injection pipeline and a micro-orifice flowmeter. The gas-liquid mixer is provided at the upper end of the simulated wellbore, with the pressure-disturbing device connected below it. Multiple pressure sensors are provided at the middle of the simulated wellbore and connected to the computer via wires and an oscilloscope.

Abstract: Embodiments are provided for building a configurable multilingual model. A computing system obtains a plurality of language-specific automatic speech recognition modules and a universal automatic speech recognition module trained on a multi-language training dataset comprising training data corresponding to each of the plurality of different languages. The computing system then compiles the universal automatic speech recognition module with the plurality of language-specific automatic speech recognition modules to generate a configurable multilingual model that is configured to selectively and dynamically utilize a sub-set of the plurality of language-specific automatic speech recognition modules with the universal automatic speech recognition module to process audio content in response to user input identifying one or more target languages associated with the audio content.

Abstract: A device and a method for early monitoring of gas intrusion based on pressure wave propagation are provided in the present disclosure. The technical solution is as follows. The lower end of the liquid storage tank is connected to the simulated wellbore through a liquid injection pipeline, a centrifugal pump, a pressure-stabilizing water tank, and a mass flowmeter. One end of the gas storage tank is connected to a screw air compressor, while the other end is connected to the simulated wellbore through an injection pipeline and a micro-orifice flowmeter. The gas-liquid mixer is provided at the upper end of the simulated wellbore, with the pressure-disturbing device connected below it. Multiple pressure sensors are provided at the middle of the simulated wellbore and connected to the computer via wires and an oscilloscope.

Abstract: The computing system trains an end-to-end (E2E) automatic speech recognition (ASR) model, using a transformer-transducer-based deep neural network that comprises a transformer encoder network and a transducer predictor network. The E2E ASR model is trained to have one or more adjustable hyperparameters that are configured to dynamically adjust an efficiency or a performance of the E2E ASR model when the E2E ASR model is deployed onto a device or executed by the device, by identifying one or more conditions of the device associated with computational power of the device and setting at least one of the one or more adjustable hyperparameters based on one or more conditions of the device.

Abstract: A video delivery system generates and stores reduced bandwidth videos from source video. The system may include a track generator that executes functionality of application(s) to be used at sink devices, in which the track generator generates tracks from execution of the application(s) on source video and generates tracks having a reduced data size as compared to the source video. The track generator may execute a first instance of application functionality on the source video, which identifies region(s) of interest from the source video. The track generator further may downsample the source video according to downsampling parameters, and execute a second instance of application functionality on the downsampled video. The track generator may determine, from a comparison of outputs from the first and second instances of the application, whether the output from the second instance of application functionality is within an error tolerance of the output from the first instance of application functionality.

Abstract: A video classification, indexing, and retrieval system is disclosed that classifies and retrieves video along multiple indexing dimensions. A search system may field queries identifying desired parameters of video, search an indexed database for videos that match the query parameters, and create clips extracted from responsive videos that are provided in response. In this manner, different queries may cause different clips to be created from a single video, each clip tailored to the parameters of the query that is received.

Abstract: Systems and methods are provided for training a machine learning model to learn speech representations. Labeled speech data or both labeled and unlabeled data sets is applied to a feature extractor of a machine learning model to generate latent speech representations. The latent speech representations are applied to a quantizer to generate quantized latent speech representations and to a transformer context network to generate contextual representations. Each contextual representation included in the contextual representations is aligned with a phoneme label to generate phonetically-aware contextual representations. Quantized latent representations are aligned with phoneme labels to generate phonetically aware latent speech representations.

Abstract: The computing system trains an end-to-end (E2E) automatic speech recognition (ASR) model, using a transformer-transducer-based deep neural network that comprises a transformer encoder network and a transducer predictor network. The E2E ASR model is trained to have one or more adjustable hyperparameters that are configured to dynamically adjust an efficiency or a performance of the E2E ASR model when the E2E ASR model is deployed onto a device or executed by the device, by identifying one or more conditions of the device associated with computational power of the device and setting at least one of the one or more adjustable hyperparameters based on one or more conditions of the device.

Abstract: Systems and methods are provided for training a machine learning model to learn speech representations. Labeled speech data or both labeled and unlabeled data sets is applied to a feature extractor of a machine learning model to generate latent speech representations. The latent speech representations are applied to a quantizer to generate quantized latent speech representations and to a transformer context network to generate contextual representations. Each contextual representation included in the contextual representations is aligned with a phoneme label to generate phonetically-aware contextual representations. Quantized latent representations are aligned with phoneme labels to generate phonetically aware latent speech representations.

Abstract: A computing system is configured to generate a transformer-transducer-based deep neural network. The transformer-transducer-based deep neural network comprises a transformer encoder network and a transducer predictor network. The transformer encoder network has a plurality of layers, each of which includes a multi-head attention network sublayer and a feed-forward network sublayer. The computing system trains an end-to-end (E2E) automatic speech recognition (ASR) model, using the transformer-transducer-based deep neural network. The E2E ASR model has one or more adjustable hyperparameters that are configured to dynamically adjust an efficiency or a performance of E2E ASR model when the E2E ASR model is deployed onto a device or executed by the device.

Abstract: A method for preparing an array substrate, a display panel and an evaporation apparatus are disclosed. A method comprises: fixing a base substrate to an evaporation stage; attaching at least one shielding sheet to the base substrate to cover at least a preset area of the base substrate; arranging and aligning an open mask in association with the base substrate, wherein the open mask has at least one opening for vapor deposition, and the at least one shielding sheet is positioned corresponding to the at least one opening and each has an area that is less than an area of a corresponding opening to shield a portion of the corresponding opening, and wherein the portion of the corresponding opening is separate from a boundary of the corresponding opening; and evaporating to form an evaporation material layer on the base substrate, to which the shielding sheet is attached.

Abstract: In an example method, a system accesses first input data and a machine learning architecture. The machine learning architecture includes a first module having a first neural network, a second module having a second neural network, and a third module having a third neural network. The system generates a first feature set representing a first portion of the first input data using the first neural network, and a second feature set representing a second portion of the first input data using the second neural network. The system generates, using the third neural network, first output data based on the first feature set and the second feature set.

Abstract: A single-tube double-wellhead drilling and production apparatus is provided, which includes a suction pile, two first wellhead heads, two second wellhead heads, two second guiding pipes, two tubings, two tubing hangers, and two plugs located inside and connected with a corresponding second wellhead head. Each second wellhead head is connected to a corresponding first wellhead head and is connected through a connecting pipeline which is provided with a control valve and a check valve. One end of each first wellhead head is connected with a corresponding first guiding pipe connected with the suction pile. Each second guiding pipe is located inside a corresponding first guiding pipe and is connected to an end of a corresponding second wellhead head. Each tubing hanger is located in and connected with a corresponding second wellhead head, is located below and communicates with a corresponding plug, end of which is provided with a corresponding tubing.

Abstract: Techniques disclosed for managing video captured by an imaging device. Methods disclosed capture a video in response to a capture command received at the imaging device. Following a video capture, techniques for classifying the captured video based on feature(s) extracted therefrom, for marking the captured video based on the classification, and for generating a media item from the captured video according to the marking are disclosed. Accordingly, the captured video may be classified as representing a static event, and, as a result, a media item of a still image may be generated. Otherwise, the captured video may be classified as representing a dynamic event, and, as a result, a media item of a video may be generated.

Abstract: A method for preparing an array substrate, a display panel and an evaporation apparatus are disclosed. A method comprises: fixing a base substrate to an evaporation stage; attaching at least one shielding sheet to the base substrate to cover at least a preset area of the base substrate; arranging and aligning an open mask in association with the base substrate, wherein the open mask has at least one opening for vapor deposition, and the at least one shielding sheet is positioned corresponding to the at least one opening and each has an area that is less than an area of a corresponding opening to shield a portion of the corresponding opening, and wherein the portion of the corresponding opening is separate from a boundary of the corresponding opening; and evaporating to form an evaporation material layer on the base substrate, to which the shielding sheet is attached.

Abstract: Systems and methods are provided for training a machine learning model to learn speech representations. Labeled speech data or both labeled and unlabeled data sets is applied to a feature extractor of a machine learning model to generate latent speech representations. The latent speech representations are applied to a quantizer to generate quantized latent speech representations and to a transformer context network to generate contextual representations. Each contextual representation included in the contextual representations is aligned with a phoneme label to generate phonetically-aware contextual representations. Quantized latent representations are aligned with phoneme labels to generate phonetically aware latent speech representations.

Abstract: The present invention relates to the field of display technology, and discloses a method for preparing an array substrate, a display panel and an evaporation apparatus. A method for preparing an array substrate comprises: fixing a base substrate to an evaporation stage; attaching a shielding sheet to the base substrate to cover at least a preset area of the base substrate; arranging and aligning an open mask in association with the base substrate; and evaporating to form a evaporation material layer on the base substrate, to which the shielding sheet is attached, with the open mask.