Abstract: Representative embodiments disclose mechanisms to separate and recognize multiple audio sources (e.g., picking out individual speakers) in an environment where they overlap and interfere with each other. The architecture uses a microphone array to spatially separate out the audio signals. The spatially filtered signals are then input into a plurality of separators, so each signal is input into a corresponding signal. The separators use neural networks to separate out audio sources. The separators typically produce multiple output signals for the single input signals. A post selection processor then assesses the separator outputs to pick the signals with the highest quality output. These signals can be used in a variety of systems such as speech recognition, meeting transcription and enhancement, hearing aids, music information retrieval, speech enhancement and so forth.

Abstract: In some embodiments, the disclosed subject matter involves a system and method relating to using an ambient capture device including a fisheye camera and a microphone array to capture audio and video in an environment, for use in an artificial intelligence (Al) application. The device with fisheye camera may provide approximately a 360° audio and video view, at relatively low cost. An embodiment may utilize a speech and vision fusion model component. The speech and vision fusion model may be trained using deep learning to combine features from many different sources, including available sensor data from the capture device. A long short term memory (LSTM) model may inter or identify features such as, but not limited to: audio direction; vision detection and tracking; voice signature; facial signature; gesture recognition; and object identification. The fusion processing may be performed by a cloud server, enabling the capture device to remain less complex.

Abstract: Representative embodiments disclose mechanisms to separate and recognize multiple audio sources (e.g., picking out individual speakers) in an environment where they overlap and interfere with each other. The architecture uses a microphone array to spatially separate out the audio signals. The spatially filtered signals are then input into a plurality of separators, so each signal is input into a corresponding signal. The separators use neural networks to separate out audio sources. The separators typically produce multiple output signals for the single input signals. A post selection processor then assesses the separator outputs to pick the signals with the highest quality output. These signals can be used in a variety of systems such as speech recognition, meeting transcription and enhancement, hearing aids, music information retrieval, speech enhancement and so forth.

Abstract: Statistical Machine Translation (SMT) based search query spelling correction techniques are described herein. In one or more implementations, search data regarding searches performed by clients may be logged. The logged data includes query correction pairs that may be used to ascertain error patterns indicating how misspelled substrings may be translated to corrected substrings. The error patterns may be used to determine suggestions for an input query and to develop query correction models used to translate the input query to a corrected query. In one or more implementations, probabilistic features from multiple query correction models are combined to score different correction candidates. One or more top scoring correction candidates may then be exposed as suggestions for selection by a user and/or provided to a search engine to conduct a corresponding search using the corrected query version(s).

Abstract: Techniques and architectures may be used to generate and perform a process using weighted finite-state transducers involving generic input search graphs. The process need not pursue theoretical optimality and instead search graphs may be optimized without an a priori optimization step. The process may result in an automatic speech recognition (ASR) decoder that is substantially faster than ASR decoders the include the optimization step.

Abstract: Techniques and architectures may be used to generate and perform a process using weighted finite-state transducers involving generic input search graphs. The process need not pursue theoretical optimality and instead search graphs may be optimized without an a priori optimization step. The process may result in an automatic speech recognition (ASR) decoder that is substantially faster than ASR decoders the include the optimization step.

Abstract: The subject disclosure is directed towards developing a translation model for mapping search query terms to document-related data. By processing user logs comprising search histories into word-aligned query-document pairs, the translation model may be trained using data, such as probabilities, corresponding to the word-aligned query-document pairs. After incorporating the translation model into model data for a search engine, the translation model is used may used as features for producing relevance scores for current search queries and ranking documents/advertisements according to relevance.

Abstract: A log-linear model may be trained using a modified version of an original limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) algorithm. The modified version may be based on modifying the original L-BFGS algorithm using a single map-reduce implementation. In another aspect, a sparse log-linear model may be accessed. The sparse log-linear model may be trained with L1-regularization, based on data indicating past user ad selection behaviors. A probability of a user selection of an ad may be determined based on the sparse log-linear model.

Abstract: Computer-readable media, computer systems, and computing methods are provided for recommending websites that are relevant to a current website to which a user has navigated. A search engine is used to track a set of websites the user has visited immediately prior to the current website, while predictive model(s) are used to generate a sequence of websites that include the current website and the tracked websites. The sequence is compared against strings of websites within a browser-history log to identify matching strings, where the matching strings include the sequence and a respective candidate website. A probability of relevance is computed from a frequency that each of the matching strings has been visited within a predefined time frame. The probability of relevance for each of the matching strings is ranked against one another to distill the highest-ranked matching strings, which are parsed to extract and present the candidate websites included therein.

Abstract: Statistical Machine Translation (SMT) based search query spelling correction techniques are described herein. In one or more implementations, search data regarding searches performed by clients may be logged. The logged data includes query correction pairs that may be used to ascertain error patterns indicating how misspelled substrings may be translated to corrected substrings. The error patterns may be used to determine suggestions for an input query and to develop query correction models used to translate the input query to a corrected query. In one or more implementations, probabilistic features from multiple query correction models are combined to score different correction candidates. One or more top scoring correction candidates may then be exposed as suggestions for selection by a user and/or provided to a search engine to conduct a corresponding search using the corrected query version(s).

Abstract: The subject disclosure is directed towards developing a translation model for mapping search query terms to document-related data. By processing user logs comprising search histories into word-aligned query-document pairs, the translation model may be trained using data, such as probabilities, corresponding to the word-aligned query-document pairs. After incorporating the translation model into model data for a search engine, the translation model is used may used as features for producing relevance scores for current search queries and ranking documents/advertisements according to relevance.

Abstract: Methods, computer systems and computer readable media for optimizing non-deterministic computational paths are provided. In embodiments, requests are received to generate reports derived from a plurality of series of data files whose metadata attributes form certain mathematical structures that can be used to choose the optimal path in the non-deterministic dependency model. Storage for each of the series of data files is optimized. Available data files needed for the report are processed and missing data files are identified. Based on the mathematical structure of the plurality of series of data files, an optimal transition with the missing data files available is determined. An entry into the transition is triggered and the missing data files are processed. The report is generated and the optimized storage is retained for future requests.

Abstract: Systems, methods, and computer media for efficiently processing user log data are provided. A received user log data analysis request specifies: target user log features that identify users in a target user group, analysis user log features that identify data associated with the users in the target user group, and an analysis to perform on the identified data associated with the users in the target user group. Occurrences of specified features are extracted from user logs and stored. Users associated with an occurrence of each of the extracted and stored target user log features are identified as users in the target user group. Occurrences of the analysis user log features that are associated with a user in the target user group are extracted and reformatted for the analysis specified in the analysis request.

Abstract: Computer-readable media, computer systems, and computing methods are provided for recommending websites that are relevant to a current website to which a user has navigated. A search engine is used to track a set of websites the user has visited immediately prior to the current website, while predictive model(s) are used to generate a sequence of websites that include the current website and the tracked websites. The sequence is compared against strings of websites within a browser-history log to identify matching strings, where the matching strings include the sequence and a respective candidate website. A probability of relevance is computed from a frequency that each of the matching strings has been visited within a predefined time frame. The probability of relevance for each of the matching strings is ranked against one another to distill the highest-ranked matching strings, which are parsed to extract and present the candidate websites included therein.

Abstract: A concrete photovoltaic structure includes an essentially vertical support member constructed, at least in part, of concrete, wherein the essentially vertical support member includes a first portion configured to be received, at least in part, within a footing. The concrete photovoltaic structure further includes a photovoltaic support member constructed, at least in part, of concrete, wherein the photovoltaic support member is configured to be affixed to a second portion of the essentially vertical support member.

Abstract: An apparatus, method and computer-readable medium provide for detecting and filtering potential race conditions. In one example, potential race conditions may be detected by determining overlap of memory accesses (e.g., read, write). Potential race conditions may be filtered as potentially benign or harmful race conditions using, for example, heuristics, comparison of output states from execution of instructions or operations in differing sequences or identifying of associated output bugs corresponding to the execution of operations in different sequences.

Abstract: A rechargeable zinc cell with a longitudinally-folded separator comprising a zinc negative electrode, a positive electrode, an electrolyte and a separator. The separator comprises at least two wicking layers with a microporous layer in the center thereof, and the separator is folded longitudinally to wrap around a long edge of the zinc negative electrode. A method of constructing a rechargeable zinc cell with a longitudinally-folded separator comprising the steps of placing the zinc negative electrode in contact with at least one of the two wicking layers of the separator, folding the separator longitudinally around a long edge of the zinc negative electrode, placing the positive electrode on said separator and rolling the zinc negative electrode, the positive electrode and the separator into a jelly roll structure.

Abstract: Methods, systems, and computer-readable media having computer-executable instructions embodied thereon that, when executed, perform methods in accordance with embodiments hereof, for emulating behavior of a target program to identify defects therein. Emulation includes recording a trace file upon running machine-level instructions of the target program, collecting relevant events encountered upon replaying the trace file, expressing the relevant events as symbolic constraints, and solving the symbolic constraints for variant input parameters. Injecting the variant input parameters into the trace file causes the target program to follow various control paths, allowing for systematically searching the target program for defects. The procedure for security testing above can be repeated by employing a search-strategy algorithm that selects optimal control paths of the target program to evaluate.

Abstract: An apparatus, method and computer-readable medium provide for detecting and filtering potential race conditions. In one example, potential race conditions may be detected by determining overlap of memory accesses (e.g., read, write). Potential race conditions may be filtered as potentially benign or harmful race conditions using, for example, heuristics, comparison of output states from execution of instructions or operations in differing sequences or identifying of associated output bugs corresponding to the execution of operations in different sequences.

Abstract: An execution of a software program can be analyzed to detect various conditions, such as software defects relating to pointers and the like. Analysis can include modeling software constructs such as heaps, calls, memory, threads, and the like. Additional information, such as call stacks, can be provided to assist in debugging. A graphical depiction of pointer history can be presented and used to navigate throughout the execution history of a program.