Abstract: A method is provided for generating a classification model configured to select an optimal execution combination for query processing. The method includes providing training queries and different execution combinations for executing the training queries. Each different execution combination involves a respective different query engine and a respective different runtime. The method includes extracting, using Cost-Based Optimizers (CBOs), a set of feature vectors for each training query. The method includes merging the set of feature vectors for the each of the training queries into a respective merged feature vector to obtain a set of merged feature vectors. The method includes adding, to each of the merged feature vectors, a respective label indicative of the optimal execution combination based on actual respective execution times of the different execution combinations, to obtain a set of labels. The method includes training the classification model by learning the merged feature vectors with the labels.

Abstract: A method for audio recognition comprises: dividing audio data to be recognized to obtain a plurality of frames of audio data; calculating, based on audio variation trends among the plurality of frames and within each of the plurality of frames, a characteristic value for each frame of the audio data to be recognized; and matching the characteristic value of each frame of the audio data to be recognized with a pre-established audio characteristic value comparison table to obtain a recognition result, wherein the audio characteristic value comparison table is established based on the audio variation trends among the frames and within each of the frames of sample data.

Abstract: Systems and processes for operating an electronic device to train a machine-learning translation system are described. In one process, a first set of training data is obtained. The first set of training data includes at least one payload in a first language and a translation of the at least one payload in a second language. The process further includes obtaining one or more templates for adapting the at least one payload; adapting the at least one payload using the one or more templates to generate at least one adapted payload formulated as a translation request; generating a second set of training data based on the at least one adapted payload; and training the machine-learning translation system using the second set of training data.

Abstract: A system determines intent values based on an object in a received phrase, and detail values based on the object in the received phrase. The system determines intent state values based on the intent values and the detail values, and detail state values and an intent detail value based on the intent values and the detail values. The system determines other intent values based on the intent values and another object in the received phrase, and other detail values based on the detail values and the other object in the received phrase. The system determines a general intent value based on the other intent values, the other detail values, and the intent state values, and another intent detail value based on the other intent values, the other detail values, and the detail state values.

Abstract: A system for assisting sharing of information includes circuitry to: input a plurality of sentences each representing a statement made by one of a plurality of users, the sentence being generated by speaking or writing during a meeting or by extracting from at least one of meeting data, email data, electronic file data, and chat data at any time; determine a statement type of the statement represented by each one of the plurality of sentences, the statement type being one of a plurality of statement types previously determined; select, from among the plurality of sentences being input, one or more sentences each representing a statement of a specific statement type of the plurality of types; and output a list of the selected one or more sentences as key statements of the plurality of sentences.

Abstract: Methods and systems for providing a correct voice command. One system includes a communication device that includes an electronic processor configured to receive a first voice command via a microphone and analyze the first voice command using a first type of voice recognition. The electronic processor determines that an action to be performed in accordance with the first voice command is unrecognizable based on the analysis using the first type of voice recognition. The electronic processor transmits the first voice command to a remote electronic computing device accompanying a request requesting that the first voice command be analyzed using a second type of voice recognition different from the first type of voice recognition. The electronic processor receives, from the remote electronic computing device, a second voice command corresponding to the action and different from the first voice command, and outputs, with a speaker, the second voice command.

Abstract: A service for generating textual transcriptions of video content is provided. A textual output generation service utilize machine learning techniques provide additional context for textual transcription. The textual output generation service first utilizes a machine learning algorithm to analyze video data from the video content and identify a set of context keywords corresponding to items identified in the video data. The textual output generation service then identifies one or more custom dictionaries of relevant terms based on the identified keywords. The textual output generation service can then utilize a machine learning algorithm to process the audio data from the video content biased with the selected dictionaries. The processing result can be utilized used to generate closed captioning information, textual content streams or otherwise stored.

Abstract: There is provided an information processing apparatus, information processing method, and program that make it possible to appropriately determine the cluster segment of the character string group that is specified on the basis of the speech recognition of the collected speech. The information processing apparatus includes: an acquisition unit that acquires a detection result relating to a variation of a sound attribute of a collected speech; and a determination unit that determines, on the basis of the detection result, a cluster segment relating to a character string group that is specified on the basis of speech recognition of the speech.

Abstract: A word vector processing method is provided. Word segmentation is performed on a corpus to obtain words, and n-gram strokes corresponding to the words are determined. Each n-gram stroke represents n successive strokes of a corresponding word. Word vectors of the words and stroke vectors of the n-gram strokes are initialized corresponding to the words. After performing the word segmentation, the n-gram strokes are determined, and the word vectors and stroke vectors are determined, training the word vectors and the stroke vectors.

Abstract: Systems and methods for distributed voice processing are disclosed herein. In one example, the method includes detecting sound via a microphone array of a first playback device and analyzing, via a first wake-word engine of the first playback device, the detected sound. The first playback device may transmit data associated with the detected sound to a second playback device over a local area network. A second wake-word engine of the second playback device may analyze the transmitted data associated with the detected sound. The method may further include identifying that the detected sound contains either a first wake word or a second wake word based on the analysis via the first and second wake-word engines, respectively. Based on the identification, sound data corresponding to the detected sound may be transmitted over a wide area network to a remote computing device associated with a particular voice assistant service.

Abstract: A system and method for processing digital multimedia files to provide searchable results includes the steps of converting a digital multimedia file to a plain text data format, annotating each word in the file with an indicator such as a time stamp to indicate where the word appears in the file, converting each indicator to an encoded indicator using characters that are not indexed by search software, indexing the converted, annotated file, storing the converted, annotated file and a file location of the converted, annotated file, receiving a query from a user's computer, and returning search results to the user's computer that include search snippets comprising unindexed portions of one or more files considered responsive to the query and the file location of those files.

Abstract: Techniques for improving quality of classification models for differentiating different user intents by improving the quality of training samples used to train the classification models are described. Pairs of user intents that are difficult to differentiate by classification models trained using the given training samples are identified based upon distinguishability scores (e.g., F-scores). For each of the identified pairs of intents, pairs of training samples each including a training sample associated with a first intent and a training sample associated with a second intent in the pair of intents are ranked based upon a similarity score between the two training samples in each pair of training samples. The identified pairs of intents and the pairs of training samples having the highest similarity scores may be presented to users through a user interface, along with user-selectable options or suggestions for improving the training samples.

Abstract: Disclosed is a method including a prior phase for referencing an authorized user, during which this user pronounces a reference phrase at least once, the phrase being converted into a series of reference symbols by a statistical conversion mutual to all of the users to be referenced, and an authentication test phase, including a first step during which a candidate user pronounces the reference phrase at least once, and this pronounced phrase is converted in the same manner as the reference phrase during the prior phase, by using the same conversion, into a sequence of candidate symbols, and a second step during which the series of candidate symbols is compared to the series of reference symbols to determine a comparison result, which is compared to at least one predetermined threshold, determining whether the candidate user who pronounced the phrase during the test phase is indeed the authorized user, providing authentication.

Abstract: A communication is received. The communication is analyzed to determine a form of the communication and a recipient of the communication. An encoded audio signal is transmitted to the recipient. Responsive to transmitting the encoded audio signal, a response encoded audio signal is received. Responsive to receiving the response encoded audio signal, the communication is transmitted to the recipient based on the response encoded audio signal.

Abstract: A speech processing system includes an input for receiving an input utterance spoken by a user and a word alignment unit configured to align different sequences of acoustic speech models with the input utterance spoken by the user. Each different sequence of acoustic speech models corresponds to a different possible utterance that a user might make. The system identifies any parts of a read prompt text that the user skipped; any parts of the read prompt text that the user repeated; and any speech sounds that the user inserted between words of the read prompt text. The information from the word alignment unit can be used to assess the proficiency and/or fluency of the user's speech.

Abstract: Acoustic change is detected by a method including preparing a first Gaussian Mixture Model (GMM) trained with first audio data of first speech sound from a speaker at a first distance from an audio interface and a second GMM generated from the first GMM using second audio data of second speech sound from the speaker at a second distance from the audio interface; calculating a first output of the first GMM and a second output of the second GMM by inputting obtained third audio data into the first GMM and the second GMM; and transmitting a notification in response to determining at least that a difference between the first output and the second output exceeds a threshold. Each Gaussian distribution of the second GMM has a mean obtained by shifting a mean of a corresponding Gaussian distribution of the first GMM by a common channel bias.

Abstract: In various embodiments, the systems and methods described herein relate to generative models. The generative models may be trained using machine learning approaches, with training sets comprising chemical compounds and biological or chemical information that relate to the chemical compounds. Deep learning architectures may be used. In various embodiments, the generative models are used to generate chemical compounds that have desired characteristics, e.g. activity against a selected target. The generative models may be used to generate chemical compounds that satisfy multiple requirements.

Abstract: Systems, methods, and devices for recognizing a user are disclosed. A speech-controlled device captures a spoken utterance, and sends audio data corresponding thereto to a server. The server determines content sources storing or having access to content responsive to the spoken utterance. The server also determines multiple users associated with a profile of the speech-controlled device. Using the audio data, the server may determine user recognition data with respect to each user indicated in the speech-controlled device's profile. The server may also receive user recognition confidence threshold data from each of the content sources. The server may determine user recognition data associated that satisfies (i.e., meets or exceeds) a most stringent (i.e., highest) of the user recognition confidence threshold data. Thereafter, the server may send data indicating a user associated with the user recognition data to all of the content sources.

Abstract: A method includes acquiring, by a camera, an image frame of an object having known geometry in a real scene, and estimating a pose of the object in the image frame with respect to the camera. A cursor is displayed on a display by rendering the cursor at a 3D position in a 3D coordinate system. An output is presented to a user when a predetermined portion of the object falls at the 3D position. The content of the output is based on the predetermined portion of the object.

Abstract: In general, a system is described that includes a set of one or more cameras and a computing device. The computing device receives a plurality of images of a three-dimensional environment captured by the one or more cameras, and a respective camera that captures a respective image is distinctly positioned at a respective particular location and in a respective particular direction. The computing device generates a plurality of image sets that each include at least three images. For each image set, the computing device calculates a plurality of predicted pairwise directions. The computing device compares a first sum of model pairwise directions with a second sum of the plurality of predicted pairwise directions and generates an inconsistency score for the respective image set. The computing device then reconstructs a digital representation of the three-dimensional environment depicted in the images.

Abstract: Techniques to dynamically customize a menu system presented to a user by a voice interaction system are provided. Audio data from a user that includes the speech of a user can be received. Features can be extracted from the received audio data, including a vocabulary of the speech of the user. The extracted features can be compared to features associated with a plurality of user group models. A user group model to assign to the user from the plurality of user group models can be determined based on the comparison. The user group models can cluster users together based on estimated characteristics of the users and can specify customized menu systems for each different user group. Audio data can then be generated and provided to the user in response to the received audio data based on the determined user group model assigned to the user.

Abstract: An apparatus for calculating acoustic score, a method of calculating acoustic score, an apparatus for speech recognition, a method of speech recognition, and an electronic device including the same are provided. An apparatus for calculating acoustic score includes a preprocessor configured to sequentially extract audio frames into windows and a score calculator configured to calculate an acoustic score of a window by using a deep neural network (DNN)-based acoustic model.

Abstract: Disclosed are machine learning-based technologies that analyze an audio input and provide speaker state predictions in response to the audio input. The speaker state predictions can be selected and customized for each of a variety of different applications.

Abstract: In one example in accordance with the present disclosure, a method for a web technology responsive to mixtures of emotions includes receiving, from a user, voice information related to the web technology. The method includes generating, using a voice analysis service, percentages or levels of different emotions detected in the voice information. The method includes activating, in the web technology, at least one of multiple defined designs or functions based on the different emotions detected. Each design or function may be activated when a particular percentage or level of an emotion is detected or when a particular mixture of different emotions is detected.

Abstract: A device, system and method for causing an output device to provide information for voice command functionality is provided. A controller determines when a received textual term, received at the controller via one or more of an input device and a communications unit, is phonetically similar to one or more existing textual terms used for activating functionality at a communication device using a voice recognition algorithm. When the received textual term, are phonetically similar to one or more existing textual terms, the controller: generates one or more suggested textual terms, related to the received textual term, that minimizes phonetic similarities with the one or more existing textual terms; and causes an output device to provide an indication of the one or more suggested textual terms to use in place of the received textual term.

Abstract: A conversational system receives an utterance, and a parser performs a parsing operation on the utterance, resulting in some words being parsed and some words not being parsed. For the words that are not parsed, words or phrases determined to be unimportant are ignored. The resulting unparsed words are processed to determine the likelihood they are important and whether they should be addressed by the automated assistant. For example, if a score associated with an important unparsed word achieves a particular threshold, then a course of action to take for the utterance may include providing a message that the portion of the utterance associated with the important unparsed word cannot be handled.

Abstract: A speaker identification system (“system”) automatically assigns a speaker to voiced segments in a conversation, without requiring any previously recorded voice sample or any other action by the speaker. The system enables unsupervised learning of speakers' fingerprints and using such fingerprints for identifying a speaker in a recording of a conversation. The system identifies one or more speakers, e.g., representatives of an organization, who are in conversation with other speakers, e.g., customers of the organization. The system processes recordings of conversations between a representative and one or more customers to generate multiple voice segments having a human voice, identifies the voice segments that have the same or a similar feature, and determines the voice in the identified voice segments as the voice of the representative.

Abstract: Embodiments described herein includes a system comprising a processor coupled to display devices, sensors, remote client devices, and computer applications. The computer applications orchestrate content of the remote client devices simultaneously across the display devices and the remote client devices, and allow simultaneous control of the display devices. The simultaneous control includes automatically detecting a gesture of at least one object from gesture data received via the sensors. The detecting comprises identifying the gesture using only the gesture data. The computer applications translate the gesture to a gesture signal, and control the display devices in response to the gesture signal.

Abstract: A system for assisting sharing of information includes circuitry to: input a plurality of sentences each representing a statement made by one of a plurality of users, the sentence being generated by speaking or writing during a meeting or by extracting from at least one of meeting data, email data, electronic file data, and chat data at any time; determine a statement type of the statement represented by each one of the plurality of sentences, the statement type being one of a plurality of statement types previously determined; select, from among the plurality of sentences being input, one or more sentences each representing a statement of a specific statement type of the plurality of types; and output a list of the selected one or more sentences as key statements of the plurality of sentences.

Abstract: Disclosed are apparatuses and methods for processing a control command for an electronic device based on a voice agent. The apparatus includes a command tagger configured to receive at least one control command for the electronic device from at least one voice agent and to tag additional information to the at least one control command, and a command executor configured to, in response to the command tagger receiving a plurality of control commands, integrate the plurality of control commands based on additional information tagged to each of the plurality of control commands and to control the electronic device based on a result of the integration.

Abstract: Systems and methods described herein utilize supervised machine learning to generate a model for scoring interview responses. The system may access a training response, which in one embodiment is an audiovisual recording of a person responding to an interview question. The training response may have an assigned human-determined score. The system may extract at least one delivery feature and at least one content feature from the audiovisual recording of the training response, and use the extracted features and the human-determined score to train a response scoring model for scoring interview responses. The response scoring model may be configured based on the training to automatically assign scores to audiovisual recordings of interview responses. The scores for interview responses may be used by interviewers to assess candidates.

Abstract: In general, the subject matter described in this specification can be embodied in methods, systems, and program products for receiving a voice query at a mobile computing device and generating data that represents content of the voice query. The data is provided to a server system. A textual query that has been determined by a speech recognizer at the server system to be a textual form of at least part of the data is received at the mobile computing device. The textual query is determined to include a carrier phrase of one or more words that is reserved by a first third-party application program installed on the computing device. The first third-party application is selected, from a group of one or more third-party applications, to receive all or a part of the textual query. All or a part of the textual query is provided to the selected first application program.

Abstract: Embodiments of the disclosure disclose a method and apparatus for outputting information. A specific embodiment of the method includes: receiving voice information, analyzing the voice information to generate voiceprint information; matching the voiceprint information with at least one piece of pre-stored voiceprint information; outputting, in response to determining the voiceprint information failing to match a piece of pre-stored voiceprint information in the at least one piece of pre-stored voiceprint information, a voice questioning message for determining whether to add a new user, and receiving a voice reply message returned from a user based on the questioning message; and outputting, in response to determining the voice reply message instructing to add the new user, a voice prompt message prompting the user to bind an account. The embodiment has improved the flexibility in the human-computer interaction.

Abstract: Techniques for optimizing training data within voice user interface (VUI) of an application under development are disclosed. A VUI feedback module synthesizes human speech of a training phrase. This phrase is presented upon a speaker which is simultaneously captured upon a microphone. A speech to text framework converts the synthesized training phrase into text (textualized training phrase). The VUI feedback module compares the textualized training phrase to the actual training phrase and generates a speech training data structure that identifies similarities or dissimilarities between the textualized training phrase and the actual training phrase. This data structure may be utilized by an application developer computing system to identify training data that is most venerable to misinterpretation when a user interacts with the VUI. The VUI may subsequently be adjusted to account for the vulnerabilities to improve operations or user experience of the VUI.

Abstract: The present disclosure relates to processing domain-specific natural language commands. An example method generally includes receiving a natural language command. A command processor compares the received natural language command to a corpus of known commands to identify a probable matching command in the corpus of known commands to the received natural language command. The corpus of known commands comprises a plurality of domain-specific commands, each of which is mapped to a domain-specific action. Based on the comparison, the command processor identifies the domain-specific action associated with the probable matching command to perform in response to the received command and executes the identified domain-specific action.

Abstract: Techniques for optimizing training data within voice user interface (VUI) of an application under development are disclosed. A VUI feedback module synthesizes human speech of a training phrase. This phrase is presented upon a speaker which is simultaneously captured upon a microphone. A speech to text framework converts the synthesized training phrase into text (textualized training phrase). The VUI feedback module compares the textualized training phrase to the actual training phrase and generates a speech training data structure that identifies similarities or dissimilarities between the textualized training phrase and the actual training phrase. This data structure may be utilized by an application developer computing system to identify training data that is most venerable to misinterpretation when a user interacts with the VUI. The VUI may subsequently be adjusted to account for the vulnerabilities to improve operations or user experience of the VUI.

Abstract: Systems and methods for analyzing a large number of textual passages are described. A computing device receives the textual passages as input and generates a Raw Pair Distance (RPD) table. The device then determines a Node table and an Node-Node Distance (NND) matrix from the RPD table. An energy reduction process is used to generate an NSPACE matrix from the NND matrix. Finally, a 3D visualizer displays aspects of the Nodes table and the NSPACE matrix to a user. The systems and methods may enable a user to quickly search and understand the text relationships within the large number of textual passages.

Abstract: [Object] To present a response to a natural sentence in a more suitable aspect even in circumstances in which a natural sentence with ambiguity can be input. [Solution] An information processing device including: an acquisition unit configured to acquire an extraction result of candidates for a response to an input which is based on first information indicating a result of natural language analysis on a natural sentence acquired as the input and second information indicating a state or a situation involved in use of a predetermined device; and a control unit configured to cause a predetermined output unit to present information indicating the candidates for the response in an aspect corresponding to the extraction result of the candidates.

Abstract: Computer-implemented methods and systems are provided for generating a distributed representation of electronic transaction data. Generation may include receiving electronic transaction data including first and second entity identifiers. Generation may also include generating an output distributed representation by iteratively updating a distributed representation using the electronic transaction data. The distributed representation may include rows corresponding to first entity identifiers and rows corresponding to second entity identifiers. An iterative update may include generating a training sample and an embedding vector using the components and the distributed representation; determining, by a neural network, a predicted category from the embedding vector; and updating the distributed representation using the predicted category and the training sample. The embodiments may also include outputting the output distributed representation to determine authorization of electronic transactions.

Abstract: Computer-implemented methods and systems are provided for generating a distributed representation of electronic transaction data. Generation may include receiving electronic transaction data including first and second entity identifiers. Generation may also include generating an output distributed representation by iteratively updating a distributed representation using the electronic transaction data. The distributed representation may include rows corresponding to first entity identifiers and rows corresponding to second entity identifiers. An iterative update may include generating a training sample and an embedding vector using the components and the distributed representation; determining, by a neural network, a predicted category from the embedding vector; and updating the distributed representation using the predicted category and the training sample. The embodiments may also include outputting the output distributed representation to determine authorization of electronic transactions.

Abstract: There is provided an apparatus including a communication unit configured to transmit information permitting a second apparatus to modify stored voice recognition information based on a relationship between the first apparatus and the second apparatus.

Abstract: A speech processing apparatus, method and non-transitory computer-readable storage medium are disclosed. A speech processing apparatus may include a memory storing instructions, and at least one processor configured to process the instructions to calculate an acoustic diversity degree value representing a degree of variation in types of sounds included in a speech signal representing a speech, on a basis of the speech signal, and compensate for a recognition feature value calculated to recognize specific attribute information from the speech signal, using the acoustic diversity degree value.

Abstract: This disclosure describes, in part, context-driven device arbitration techniques to select a speech interface device from multiple speech interface devices to provide a response to a command included in a speech utterance of a user. In some examples, the context-driven arbitration techniques may include executing multiple pipeline instances to analyze audio signals and device metadata received from each of the multiple speech interface devices which detected the speech utterance. A remote speech processing service may execute the multiple pipeline instances and analyze the audio signals and/or metadata, at various stages of the pipeline instances, to determine which speech interface device is to respond to the speech utterance.

Abstract: In particular embodiments, one or more computer-readable non-transitory storage media embody software that is operable when executed to receive an audio waveform fingerprint and a client-determined location from a client device. The received audio waveform fingerprint may be compared to a database of stored audio waveform fingerprints, each stored audio waveform fingerprint associated with an object in an object database. One or more matching audio waveform fingerprints may be found from a comparison set of audio waveform fingerprints obtained from the audio waveform fingerprint database. Location information associated with a location of the client device may be determined, and the location information may be sent to the client device. The client device may be operable to update the client-determined location based at least in part on the location information.

Abstract: A processor of a vehicle speech recognition system recognizes speech via domain-specific language and acoustic models. The processor further, in response to the acoustic model having a confidence score for recognized speech falling within a predetermined range defined relative to a confidence score for the domain-specific language model, recognizes speech via the acoustic model only.

Abstract: A classification engine classifies words into gender-biased or neutral classes and outputs classification labels and probabilities the words were correctly classified. A dictionary engine generates a set of gender-biased words based on their labels, if their probability exceeds a threshold, if they are defined within a set of definitions. For each such word, the dictionary engine parses the set of definitions to extract the most frequent meaning of that word in a part of speech; parses the set of definitions to extract a synonym for the most frequent meaning of that word in that part of speech; and parses the classification engine output to determine that synonym's classification label. A synonym is selected that is labeled in the neutral class as a replacement for the gender-biased word. A dictionary entry is generated that includes the word, the part of speech, and the synonym.

Abstract: A device includes circuitry configured to acquire detection data from at least one sensor corresponding to a motion of a finger of a user. The circuitry is also configured to set a window based on the acquired data. The circuitry is also configured to identify a gesture based on a second motion performed by the user and to control a display based on the gesture.

Abstract: A voice-controlled device may receive a voice command uttered by a user, where the voice command may request that the voice-controlled device perform an operation. The voice-controlled device and/or one or more remote computing resources may process an audio signal associated with the voice command to determine text corresponding to the voice command. The resulting user utterance may be associated with a unique identifier, which may be provided to a third party and/or third party application that is to provide information responsive to the user request. The information provided by the third party/third party application may be output to the user based at least partly on the unique identifier, without disclosing user data associated with the user.

Abstract: Technologies are disclosed for voice-controlled monitoring of computing resources in a service provider network. Verbal requests for information about the status of computing resources in a service provider network are recorded and transmitted to a voice service. The spoken utterances are translated into text and analyzed to determine the intent of the speaker and to extract relevant words by a voice service. The voice service provides a request to a serverless compute service that identifies the intent of the speaker and includes the words extracted from the verbal utterance. The serverless compute service, in turn, executes a voice command processing function to identify the resources, obtain the status information, and construct a textual reply to the request that includes the status information. The textual reply can then be provided to the voice service for conversion into spoken audio and playback to the requesting user.

Abstract: An example apparatus for detecting keywords in audio includes an audio receiver to receive audio comprising a keyword to be detected. The apparatus also includes a spike transducer to convert the audio into a plurality of spikes. The apparatus further includes a spiking neural network to receive one or more of the spikes and generate a spike corresponding to a detected keyword.