Abstract: Methods, systems, and computer-readable storage media for receiving a source domain data set including a set of source document and source label pairs, each source label corresponding to a source domain and indicating a sentiment attributed to a respective source document, receiving a target domain data set including a set of target documents absent target labels, processing documents of the source and target domains using a feature encoder of a DAS platform, to map the documents of the source and target domains to a shared feature space through feature representations, the processing including minimizing a distance between the feature representations of the source domain, and feature representations of the target domain based on a set of loss functions, providing an ensemble prediction from the processing, and providing predicted labels based on the ensemble prediction, the predicted labels being used by the sentiment classifier to classify documents from the target domain.

Abstract: Systems and methods for a speech recognition system for recognizing speech including overlapping speech by multiple speakers. The system including a hardware processor. A computer storage memory to store data along with having computer-executable instructions stored thereon that, when executed by the processor is to implement a stored speech recognition network. An input interface to receive an acoustic signal, the received acoustic signal including a mixture of speech signals by multiple speakers, wherein the multiple speakers include target speakers. An encoder network and a decoder network of the stored speech recognition network are trained to transform the received acoustic signal into a text for each target speaker. Such that the encoder network outputs a set of recognition encodings, and the decoder network uses the set of recognition encodings to output the text for each target speaker. An output interface to transmit the text for each target speaker.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating an output sequence of audio data that comprises a respective audio sample at each of a plurality of time steps. One of the methods includes, for each of the time steps: providing a current sequence of audio data as input to a convolutional subnetwork, wherein the current sequence comprises the respective audio sample at each time step that precedes the time step in the output sequence, and wherein the convolutional subnetwork is configured to process the current sequence of audio data to generate an alternative representation for the time step; and providing the alternative representation for the time step as input to an output layer, wherein the output layer is configured to: process the alternative representation to generate an output that defines a score distribution over a plurality of possible audio samples for the time step.

Abstract: Conversational image editing and enhancement techniques are described. For example, an indication of a digital image is received from a user. Aesthetic attribute scores for multiple aesthetic attributes of the image are generated. A computing device then conducts a natural language conversation with the user to edit the digital image. The computing device receives inputs from the user to refine the digital image as the natural language conversation progresses. The computing device generates natural language suggestions to edit the digital image based on the aesthetic attribute scores as part of the natural language conversation. The computing device provides feedback to the user that includes edits to the digital image based on the series of inputs. The computing device also includes as feedback natural language outputs indicating options for additional edits to the digital image based on the series of inputs and the previous edits to the digital image.

Abstract: An interactive response system combines human intelligence (HI) subsystems with artificial intelligence (AI) subsystems to facilitate overall capability of multi-channel user interfaces. The system permits imperfect AI subsystems to nonetheless lessen the burden on HI subsystems. A combined AI and HI proxy is used to implement an interactive omnichannel system, and the proxy dynamically determines how many AI and HI subsystems are to perform recognition for any particular utterance, based on factors such as confidence thresholds of the AI recognition and availability of HI resources. Furthermore the system uses information from prior recognitions to automatically build, test, predict confidence, and maintain AI models and HI models for system recognition improvements.

Abstract: Symbol prediction can be implemented using a multi-task system trained for different tasks. The tasks may include a single symbol prediction, symbol category prediction, and symbol subcategory prediction. Categories of symbols can be generated by clustering sets of training data using a clustering scheme.

Abstract: A computing system determines whether a reference audio signal contains a query. A time-frequency convolutional neural network (TFCNN) comprises a time and frequency convolutional layers and a series of additional layers, which include a bottleneck layer. The computation engine applies the TFCNN to samples of a query utterance at least through the bottleneck layer. A query feature vector comprises output values of the bottleneck layer generated when the computation engine applies the TFCNN to the samples of the query utterance. The computation engine also applies the TFCNN to samples of the reference audio signal at least through the bottleneck layer. A reference feature vector comprises output values of the bottleneck layer generated when the computation engine applies the TFCNN to the samples of the reference audio signal. The computation engine determines at least one detection score based on the query feature vector and the reference feature vector.

Abstract: Techniques for using a dynamic wakeword detection threshold are described. A device detects a wakeword in audio data using a first wakeword detection threshold value. Thereafter, the device receives audio including speech. If the device receives the audio within a predetermined duration of time after detecting the previous wakeword, the device attempts to detect a wakeword in second audio data, corresponding to the audio including the speech, using a second, lower wakeword detection threshold value.

Abstract: Data update methods, systems, and devices are disclosed. The method includes: acquiring at least a first piece of audio data of a user in a first conversation scenario and at least a second piece of audio data of the user in a second conversation scenario, performing voiceprint recognition on the first and second pieces of audio data based on voiceprint information of the user, acquiring audio feature information of the first and second pieces of audio data, and updating voiceprint information of the user according to the audio feature information.

Abstract: Disclosed is a solution for diagnosing problems from logs used in an application development environment. A random sample of log statements is collected. The log statements can be completely unstructured and/or do not conform to any natural language. The log statements are tagged with predefined classifications. A natural language processing (NLP) classifier model is trained utilizing the log statements tagged with the predefined classification. New log statements can be classified into the plurality of predefined classifications utilizing the trained NLP classifier model. From the log statements thus classified, statements having a problem classification can be identified and presented through a dashboard running in a browser. Outputs from the trained NLP classifier model can be provided as input to another trained model for automatically and quickly identifying a type of problem associated with the statements, eliminating a need to manually sift through tens or hundreds of thousands of lines of logs.

Abstract: An example apparatus for routing audio streams includes an audio receiver to receive audio from a microphone. The apparatus also includes a classifier to semantically generate a result set based on the audio. The apparatus further includes a scheduler to select a spoken language understanding (SLU) engine based on the result set. The apparatus includes a router to route the audio to the selected SLU engine.

Abstract: A centralized and robust threat assessment tool is disclosed to perform comprehensive analysis of previously-stored and subsequent communication data, activity data, and other relevant information relating to inmates within a controlled environment facility. As part of the analysis, the system detects certain keywords and key interactions with the dataset in order to identify particular criminal proclivities of the inmate. Based on the identified proclivities, the system assigns threat scores to inmate that represents a relative likelihood that the inmate will carry out or be drawn to certain threats and/or criminal activities. This analysis provides a predictive tool for assessing an inmate's ability to rehabilitate. Based on the analysis, remedial measures can be taken in order to correct an inmate's trajectory within the controlled environment and increase the likelihood of successful rehabilitation, as well as to prevent potential criminal acts.

Abstract: Received data representing speech is stored, and a trigger detection block detects a presence of data representing a trigger phrase in the received data. In response, a first part of the stored data representing at least a part of the trigger phrase is supplied to an adaptive speech enhancement block, which is trained on the first part of the stored data to derive adapted parameters for the speech enhancement block. A second part of the stored data, overlapping with the first part of the stored data, is supplied to the adaptive speech enhancement block operating with said adapted parameters, to form enhanced stored data. A second trigger phrase detection block detects the presence of data representing the trigger phrase in the enhanced stored data. In response, enhanced speech data are output from the speech enhancement block for further processing, such as speech recognition.

Abstract: A method for classifying speakers includes: receiving, by a speaker recognition system including a processor and memory, input audio including speech from a speaker; extracting, by the speaker recognition system, a plurality of speech frames containing voiced speech from the input audio; computing, by the speaker recognition system, a plurality of features for each of the speech frames of the input audio; computing, by the speaker recognition system, a plurality of recognition scores for the plurality of features; computing, by the speaker recognition system, a speaker classification result in accordance with the recognition scores; and outputting, by the speaker recognition system, the speaker classification result.

Abstract: A method at an electronic device with one or more microphones and a speaker, the electronic device configured to be responsive to any of a plurality of affordances including a voice-based affordance, includes determining background noise of an environment associated with the electronic device, and before detecting the voice-based affordance: determining whether the background noise would interfere with recognition of the hotword in voice inputs detected by the electronic device, and if so, indicating to a user to use an affordance other than the voice-based affordance.

Abstract: A method at an electronic device with one or more microphones and a speaker includes receiving a first voice input; comparing the first voice input to one or more voice models; based on the comparing, determining whether the first voice input corresponds to any of a plurality of occupants, and according to the determination, authenticating an occupant and presenting a response, or restricting functionality of the electronic device.

Abstract: Speech recognition is performed on a received utterance to determine a plurality of candidate text representations of the utterance, including a primary text representation and one or more alternative text representations. Natural language processing is performed on the primary text representation to determine a plurality of candidate actionable intents, including a primary actionable intent and one or more alternative actionable intents. A result is determined based on the primary actionable intent. The result is provided to the user. A recognition correction trigger is detected. In response to detecting the recognition correction trigger, a set of alternative intent affordances and a set of alternative text affordances are concurrently displayed.

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for entity prediction. One of the methods includes performing word segmentation on text to be predicted to obtain a plurality of words. For each word of the plurality of words, a determination is made whether the word has a pre-trained word vector. In response to determining that the word has a pre-trained word vector, the pre-trained word vector for the word is obtained. In response to determining that the word does not have a pre-trained word vector, a word vector for the word is determined based on a pre-trained stroke vector. The word vector and the pre-trained stroke vector are trained based on a text sample and a word vector model. An entity associated with the text is predicted by inputting word vectors of the plurality of words into an entity prediction model.

Abstract: A method, a system, and a computer program product for preventing initiation of a voice recognition session. The method includes monitoring at least one audio output channel for at least one audio trigger phrase that initiates a voice recognition session. The method further includes in response to detecting the at least one audio trigger phrase on the at least one audio output channel, setting a logic state of at least one output trigger detector of the at least one audio output channel to a first state. The method further includes gating a logic state of at least one input trigger detector of at least one audio input channel to the first state for a time period and preventing initiation of a voice recognition session by the at least one audio trigger phrase on the at least one audio input channel while the logic state is the first state.

Abstract: Systems and processes for providing multi-directional dialog are provided. An example method includes, at an electronic device with one or more processors, receiving a first natural-language input; determining a first intent based on the first natural-language input, identifying a first dialog flow based on the first intent, outputting a natural-language output associated with the first dialog flow, receiving a second natural-language input; determining whether the second natural-language input satisfies dialog criteria associated with the first dialog flow, and in accordance with a determination that the second natural-language input satisfies the dialog criteria, outputting a second natural-language output associated with the first dialog flow.