Abstract: Methods and systems for training and/or using a language selection model for use in determining a particular language of a spoken utterance captured in audio data. Features of the audio data can be processed using the trained language selection model to generate a predicted probability for each of N different languages, and a particular language selected based on the generated probabilities. Speech recognition results for the particular language can be utilized responsive to selecting the particular language of the spoken utterance. Many implementations are directed to training the language selection model utilizing tuple losses in lieu of traditional cross-entropy losses. Training the language selection model utilizing the tuple losses can result in more efficient training and/or can result in a more accurate and/or robust model—thereby mitigating erroneous language selections for spoken utterances.

Abstract: A method for monitoring defects of a product implemented in an electronic device obtains product data in real time and determines whether a product is defective based on the product data; when the product is defective, outputting first warning information based on the number of defects of the product which satisfy a first preset condition; obtaining a rate of defects of the product every first preset time period, and outputting second warning information based on the rate of defects of the product when the rate of defects satisfies at least one of a second, third, and fourth preset conditions; when any warning information is output, analyzing distribution of the defects of the product; and predicting at least one cause of each defect of the product according to historical maintenance data of the product and a self-learning record of the electronic device.

Abstract: Methods, systems, apparatus, including computer programs encoded on computer storage medium, to facilitate language independent-speaker verification. In one aspect, a method includes actions of receiving, by a user device, audio data representing an utterance of a user. Other actions may include providing, to a neural network stored on the user device, input data derived from the audio data and a language identifier. The neural network may be trained using speech data representing speech in different languages or dialects. The method may include additional actions of generating, based on output of the neural network, a speaker representation and determining, based on the speaker representation and a second representation, that the utterance is an utterance of the user. The method may provide the user with access to the user device based on determining that the utterance is an utterance of the user.

Abstract: Techniques are described for training and/or utilizing an end-to-end speaker diarization model. In various implementations, the model is a recurrent neural network (RNN) model, such as an RNN model that includes at least one memory layer, such as a long short-term memory (LSTM) layer. Audio features of audio data can be applied as input to an end-to-end speaker diarization model trained according to implementations disclosed herein, and the model utilized to process the audio features to generate, as direct output over the model, speaker diarization results. Further, the end-to-end speaker diarization model can be a sequence-to-sequence model, where the sequence can have variable length. Accordingly, the model can be utilized to generate speaker diarization results for any of various length audio segments.

Abstract: The invention discloses a conductive paste for forming the electrode on the surface of solar cell, which contains conductive powder, mixed glass and organic phase; wherein, the mixed glass comprises the following two types of glass components: the first type of glass is at least one selected from the tellurium glass which does not contain lead substantially and having tellurium, bismuth, lithium as the essential component; The second type of glass is at least one kind of lead silicate glass, which having lead and silicon as essential components and does not contain tellurium substantially. The invention also provides a solar cell prepared by printing the conductive paste as a surface electrode and a manufacturing method of the solar cell.

Abstract: A method comprises forming a single-crystal-like metal layer on a metal seed layer, the metal seed layer formed on a carrier wafer. The method comprises forming a first bonding layer on the single-crystal-like metal layer. The method also comprises forming a second bonding layer on a dielectric layer of a target substrate, the target substrate comprising one or more recording head subassemblies. The bonding layers may include diffusion layers or dielectric bonding layers. The method further comprises flipping and joining the carrier wafer with the target substrate such that the first and second diffusion layers are bonded and the single-crystal-like metal layer is integrated with the recording head as a near-field transducer.

Abstract: Methods and systems for training and/or using a language selection model for use in determining a particular language of a spoken utterance captured in audio data. Features of the audio data can be processed using the trained language selection model to generate a predicted probability for each of N different languages, and a particular language selected based on the generated probabilities. Speech recognition results for the particular language can be utilized responsive to selecting the particular language of the spoken utterance. Many implementations are directed to training the language selection model utilizing tuple losses in lieu of traditional cross-entropy losses. Training the language selection model utilizing the tuple losses can result in more efficient training and/or can result in a more accurate and/or robust model—thereby mitigating erroneous language selections for spoken utterances.

Abstract: Processing of acoustic features of audio data to generate one or more revised versions of the acoustic features, where each of the revised versions of the acoustic features isolates one or more utterances of a single respective human speaker. Various implementations generate the acoustic features by processing audio data using portion(s) of an automatic speech recognition system. Various implementations generate the revised acoustic features by processing the acoustic features using a mask generated by processing the acoustic features and a speaker embedding for the single human speaker using a trained voice filter model. Output generated over the trained voice filter model is processed using the automatic speech recognition system to generate a predicted text representation of the utterance(s) of the single human speaker without reconstructing the audio data.

Abstract: A method includes forming a single-crystal-like metal layer on a metal seed layer, the metal seed layer formed on a sacrificial wafer. An anchor layer is formed on the single-crystal-like metal layer. The single-crystal-like metal layer is separated from the sacrificial wafer via the anchor layer. The single-crystal-like metal layer is transported via the anchor layer to a target substrate having one or more recording head subassemblies. The single-crystal-like metal layer is joined with the recording head, the single-crystal-like metal layer being integrated with the recording head as a near-field transducer.

Abstract: Methods and systems for training and/or using a language selection model for use in determining a particular language of a spoken utterance captured in audio data. Features of the audio data can be processed using the trained language selection model to generate a predicted probability for each of N different languages, and a particular language selected based on the generated probabilities. Speech recognition results for the particular language can be utilized responsive to selecting the particular language of the spoken utterance. Many implementations are directed to training the language selection model utilizing tuple losses in lieu of traditional cross-entropy losses. Training the language selection model utilizing the tuple losses can result in more efficient training and/or can result in a more accurate and/or robust model—thereby mitigating erroneous language selections for spoken utterances.

Abstract: Techniques disclosed herein enable training and/or utilizing speaker dependent (SD) speech models which are personalizable to any user of a client device. Various implementations include personalizing a SD speech model for a target user by processing, using the SD speech model, a speaker embedding corresponding to the target user along with an instance of audio data. The SD speech model can be personalized for an additional target user by processing, using the SD speech model, an additional speaker embedding, corresponding to the additional target user, along with another instance of audio data. Additional or alternative implementations include training the SD speech model based on a speaker independent speech model using teacher student learning.

Abstract: The present disclosure provides an intelligent interaction processing method and apparatus, a device and a computer storage medium. The method comprises: performing intention recognition for a preceding feedback item already returned to the user; continuing to return a subsequent feedback item to the user based on the intention of the preceding feedback item. According to the present disclosure, it is possible to guess the user's subsequent intention based on the preceding feedback item, and continue to return the desired subsequent feedback item to the user without the user's operations, so that the present disclosure is more intelligentized and richer and simplifies the user's operations.

Abstract: A hydrate kinetic inhibitor, which is prepared by a polymerization of mercaptoethanol and N-vinylcaprolactam, is hydroxyl terminated poly(N-vinylcaprolactam) having a structure of formula (I) below, wherein n=10 to 1000. The inhibitor is a novel hydrate kinetic inhibitor, which has low effective concentration and high cloud point, and is effective when the degree of supercooling is relatively high.

Abstract: A method of forming a thin film structure involves performing one or more repetitions to form a template on a wafer. The repetitions include: depositing a layer of a template material to a first thickness T1; and ion beam milling the layer of the template material to remove thickness T2, where T2<T1, resulting in a layer of the template material with thickness T1?T2. The ion beam milling is performed at a channeling angle relative to a deposition plane of the wafer, the channeling angle defined relative to a channeling direction of a crystalline microstructure of the template material. After the repetitions, additional material is deposited on the template to form a final structure. The additional material has a same crystalline microstructure as the template material.

Abstract: Techniques disclosed herein enable training and/or utilizing speaker dependent (SD) speech models which are personalizable to any user of a client device. Various implementations include personalizing a SD speech model for a target user by processing, using the SD speech model, a speaker embedding corresponding to the target user along with an instance of audio data. The SD speech model can be personalized for an additional target user by processing, using the SD speech model, an additional speaker embedding, corresponding to the additional target user, along with another instance of audio data. Additional or alternative implementations include training the SD speech model based on a speaker independent speech model using teacher student learning.

Abstract: A drug screening device is provided. A method of determining optimal drug concentrations and efficacy in a patient using the device are provided. A method of determining effective chemotherapeutic drugs and effective concentrations thereof using the device is provided. Also, a method of determining safety and efficacy of drugs using the device is provided.

Abstract: A thin film structure (e.g., a near-field transducer), includes a first surface parallel to a substrate on which the thin film structure is deposited and two other surfaces orthogonal to the first surface. The first surface and the two other surfaces have respective first, second, and third selected plane orientations with respective first, second, and third atomic packing factors. The first, second, and third selected plane orientations are selected to maximize an average of the first, second, and third atomic packing factors.

Abstract: A clustered storage system includes a plurality of storage devices, each of which contributes a portion of its memory to form a global cache of the clustered storage system that is accessible by the plurality of storage devices. Cache metadata for accessing the global cache may be organized in a multi-layered structure. In one embodiment, multi-layered structure has a first layer first including a first address array, and the first address array include addresses pointing to a plurality of second address arrays in a second layer. Each second address array in the second layer includes addresses, each of which points to data that has been cached in the global cache.

Abstract: Techniques disclosed herein enable training and/or utilizing speaker dependent (SD) speech models which are personalizable to any user of a client device. Various implementations include personalizing a SD speech model for a target user by processing, using the SD speech model, a speaker embedding corresponding to the target user along with an instance of audio data. The SD speech model can be personalized for an additional target user by processing, using the SD speech model, an additional speaker embedding, corresponding to the additional target user, along with another instance of audio data. Additional or alternative implementations include training the SD speech model based on a speaker independent speech model using teacher student learning.

Abstract: An interconnected computer system includes a Peripheral Component Interconnect Express (PCIe) fabric, a first computer system communicatively coupled to the PCIe fabric, a second computer system communicatively coupled to the PCIe fabric, and a shared single-access hardware resource coupled to the PCIe fabric. The first computer system includes a first processor and first memory coupled to the first processor configured to store a first flag indicating a desire of the first computer system to access the shared single-access hardware resource and a turn variable indicating which of the first computer system and the second computer system has access to the shared single-access hardware resource. The second computer system includes a second processor and second memory coupled to the second processor configured to store a second flag indicating a desire of the second computer system to access the shared single-access hardware resource.