Abstract: Provided is a method for determining output data for a plurality of text documents, including the steps of: providing a feature matrix as input data; wherein the feature matrix includes information about frequencies of a plurality of features within the plurality of text documents; clustering the feature matrix using a clustering algorithm into at least one clustering matrix; wherein the at least one clustering matrix includes information about the cluster membership of each document of the plurality of documents or each feature of the plurality of features, assigning at least one score to each feature of the plurality of features based on the at least one clustering matrix; ranking the plurality of features based on their assigned scores; and outputting the ranked features as output data. A corresponding computer program product and system is also provided.

Abstract: A sign language translation system may capture infrared images of the formation of a sign language sign or sequence of signs. The captured infrared images may be used to produce skeletal joints data that includes a temporal sequence of 3D coordinates of skeletal joints of hands and forearms that produced the sign language sign(s). A hierarchical bidirectional recurrent neural network may be used to translate the skeletal joints data into a word or sentence of a spoken language. End-to-end sentence translation may be performed using a probabilistic connectionist temporal classification based approach that may not require pre-segmentation of the sequence of signs or post-processing of the translated sentence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining voice characteristics are provided. One of the methods includes: obtaining speech data of a speaker; inputting the speech data into a model trained at least by jointly minimizing a first loss function and a second loss function, wherein the first loss function comprises a non-sampling-based loss function and the second loss function comprises a Gaussian mixture loss function with non-unit multi-variant covariance matrix; and obtaining from the trained model one or more voice characteristics of the speaker.

Abstract: Methods, systems, and computer programs are presented for estimating the language used in a user communication. One method includes an operation for utilizing counters to track use of languages by a user of an online service, the counters being updated based on interactions of the user in the online service. Further, the method includes operations for detecting a text entered by the user and obtaining, by a language classifier, an initial prediction having probabilities for the languages that the text is in the language. A language distribution prediction is calculated based on the initial prediction and the user counters, where the language distribution prediction comprises a probability, for each language, that the text is in the language. Further, the method includes operations for selecting a language used in the text based on the language distribution prediction and causing presentation on a display of a message in the selected language.

Abstract: An electronic device is provided. The electronic device includes a memory storing recording data including a content of a conversation and at least one instruction, and a processor configured, by executing the at least one instruction, to input first data corresponding to a first voice in the content of the conversation into a first neural network model and acquire category information of the first data, and acquire category information of second data corresponding to a second voice in the content of the conversation. The processor is configured to, based on the category information of the first data and the category information of the second data being different, train the first neural network model based on the category information of the second data and the first data.

Abstract: A translation device includes: a controller that extracts a proper noun candidate from an original sentence, generates a translation word of the proper noun candidate in a second language, generates a second translated sentence by translating the original sentence into the second language based on the proper noun candidate and the translation word of the proper noun candidate, and generates a second reverse-translated sentence by translating the second translated sentence into the first language based on the proper noun candidate and the translation word of the proper noun candidate; a display that displays the first reverse-translated sentence and the second reverse-translated sentence; and an operation unit that receives a user operation of selecting one of the first reverse-translated sentence and the second reverse-translated sentence.

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a pattern program using a genetic algorithm. The genetic algorithm operates on example data strings that represent the data categories to be recognized or extracted through named entity recognition. In the initialization stage, the initial pattern programs are generated based on example data strings that represent the data categories to be recognized or extracted through named entity recognition. Starting from the initial pattern programs, genetic operations are iteratively conducted to generate generations of offspring pattern programs. In each round of the genetic operation, offspring pattern programs are generated through the crossover operation and the mutation operation.

Abstract: Systems and techniques for generating natural language understanding (NLU) models are described. A developer of an NLU model may provide data representing runtime NLU functionality. For example, a developer may provide one or more sample natural language user inputs. The NLU model generation system may expand data, provided by the developer, to result in a more robust NLU model for use at runtime. For example, the NLU model generation system may expand sample natural language user inputs, may translate sample natural language user inputs into other languages, etc. The present disclosure also provides a mechanism for transitioning between using NLU models of a first NLU model generation system and NLU models of a second NLU model generation system.

Abstract: A display system includes a display device and a server. The server receives a plurality of audio segments from a plurality of audio-capture devices. The server receives a user-input that corresponds to a selection of a first user interface (UI) element that represents a first performer-of-interest or a first audio-capture device attached to the first performer-of-interest. The server detects a second performer-of-interest associated with a second audio-capture device within a threshold range of the first audio-capture device. The server extracts a first audio segment of a first vocal deliverance of the first performer-of-interest and a second audio segment of a second vocal deliverance of the second performer-of-interest. The server deduces new caption information from a first verbatim text that is common between the first audio segment and the second audio segment and controls display of the new caption information on the display device.

Abstract: A method of predicting a person's interests is provided. The method includes receiving geolocation information about a user location, reading, from a database of interpretations, at least one interpretation of an expression made in close proximity to the location, reading, from a database of ad bids, a plurality of ad bids comprising interpretations, comparing the interpretation from the database to the interpretations of the ad bids to select a most valuable ad bid having an interpretation that matches the interpretation of an expression made in close proximity to the location, and presenting an ad associated with the most valuable ad bid, wherein the interpretation is from a natural language expression.

Abstract: Techniques are described herein for multi-modal interaction between users, automated assistants, and other computing services. In various implementations, a user may engage with the automated assistant in order to further engage with a third party computing service. In some implementations, the user may advance through dialog state machines associated with third party computing service using both verbal input modalities and input modalities other than verbal modalities, such as visual/tactile modalities.

Abstract: A method of providing examples to a computerized conversation agent includes associating one or more triplets of two concepts and a relationship therebetween with an intent related to a query. The triplet is embedded in a knowledge graph and the concepts in the knowledge graph are mapped to the intent. A call log is scanned for examples of the intent based on the concepts in the knowledge graph and the examples are mapped to the intent.

Abstract: Interactive natural language processing (NLP) results may be generated that allow a user to interact with the NLP results but do so in an offline manner so that the documents being processed need not be stored online. To provide interactive NLP results, event handlers may be attached to elements of the NLP results. A user may then select a word or phrase of the NLP results to cause computer software provided with the NLP to present the interactive features. For example, a user may click on a definite noun phrase to view information for diagnosing antecedent basis errors. For another example, a user may click on a word to view information about how that word is used in a document, such as viewing portions of the document that include the word or variants of the word.

Abstract: A system that can engage in a dialog with a user may select a system response to a user input based on how the system estimates a user may respond to a potential system response. Models may be trained to evaluate a potential system response in view of various available data including dialog history, entity data, etc. Each model may score the potential system response for various qualitative aspects such as whether the response is likely to be comprehensible, on-topic, interesting, likely to lead to the dialog continuing, etc. Such scores may be combined to other scores such as whether the potential response is coherent or engaging. The models may be trained using previous dialog/chatbot evaluation data. At runtime the scores may be used to select a system response to a user input as part of the dialog.

Abstract: A speech translation apparatus includes: an estimator which estimates a sound source direction, based on an acoustic signal obtained by a microphone array unit; a controller which identifies that an utterer is a user or a conversation partner, based on the sound source direction estimated after the start of translation is instructed by a button, using a positional relationship indicated by a layout information item stored in storage and selected in advance, and determines a translation direction indicating input and output languages in and into which content of the acoustic signal is recognized and translated, respectively; and a translator which obtains, according to the translation direction, original text indicating the content in the input language and translated text indicating the content in the output language. The controller displays the original and translated texts on first and second display areas corresponding to the positions of the user and conversation partner, respectively.

Abstract: Techniques are described herein for multi-modal interaction between users, automated assistants, and other computing services. In various implementations, a user may engage with the automated assistant in order to further engage with a third party computing service. In some implementations, the user may advance through dialog state machines associated with third party computing service using both verbal input modalities and input modalities other than verbal modalities, such as visual/tactile modalities.

Abstract: The present disclosure proposes a federated speaker verification method based on differential privacy, including: 1. performing, by a server, UBM pre-training to obtain an initial UBM; 2. receiving, by the client, the pre-trained initial UBM, and performing initial UBM learning based on local private speech data; 3. performing, by the client, differential privacy protection based on learned statistics; 4. aggregating, by the server, statistics uploaded by multiple clients, and updating the initial UBM; and 5. receiving, by the client, the updated UBM, performing adjustment based on the local private speech data to obtain a GMM for a user of the client, and determining, based on the updated UBM and the GMM, whether a to-be-verified speech is generated by the user of the client.

Abstract: Embodiments of the present disclosure set forth a method of decomposing an audio signal into a set of sub-band signals and detecting a set of signal energy values, where each signal energy value is associated with a sub-band signal. The method also includes generating a noise reduction threshold based on at least one sub-band signal, and, for each sub-band signal, comparing the associated signal energy value to the noise reduction threshold. Based on determining that at least one sub-band signal is associated with a signal energy value below the noise reduction threshold, the method includes attenuating the at least one the sub-band signal to generate a set of attenuated sub-band signals. The method also includes combining at least one sub-band signal included in the set of sub-band signals with at least one attenuated sub-band signal included in the set of attenuated sub-band signals to generate an output audio signal.

Abstract: To suitably regulate sound pressure of object content on a receiving side. An audio stream including coded data of a predetermined number of pieces of object content is generated. A container of a predetermined format including the audio stream is transmitted. Information indicating a range within which sound pressure is allowed to increase and decrease for each piece of object content is inserted into a layer of the audio stream and/or a layer of the container. On a receiving side, sound pressure of each piece of object content increases and decreases within the allowable range based on the information.

Abstract: A method and a systems of interchanging code-mixed words and uni-language words are described. In an embodiment, the method may include identifying two or more portions of a target word, wherein the target word is one of a code-mixed word and a uni-language word. The method may further include determining one or more equivalent words corresponding to at least one portion of the two or more portions, wherein the one or more equivalent words are in at least one predetermined target language. The method may further include selecting at least one equivalent word from the one or more equivalent words, based on a context of the target word within a sentence comprising the target word. The method may further include replacing the at least one portion of the two or more portions of the target word with the selected at least one equivalent word.