Abstract: In one example embodiment, audio characteristics of audio signals are adjusted by a first machine learning model to reduce effects of a facial covering and produce adjusted audio signals. The audio signals correspond to resulting voice signals produced from the facial covering affecting original voice signals. Speech characteristics are predicted for the adjusted audio signals by a second machine learning model. Transformed audio signals corresponding to the original voice signals are produced based on the adjusted audio signals and predicted speech characteristics.

Abstract: Techniques for machine learning (ML) model adaptation via segment replacement and student-teacher training are described. A model optimizer determines that a ML model sought to be deployed to an edge device includes a source segment that is not supported by the edge device. The model optimizer identifies a replacement segment that is equivalent to the source segment, constructs an adapted ML model by swapping in the replacement segment for the source segment, and trains the adapted ML model, optionally from the source ML model using a student-teacher training procedure. The trained adapted ML model is then deployed to the edge device, where it can be successfully run with minimal (if any) degradation of performance compared to the original source ML model.

Abstract: There is provided a computer implemented method for automated analysis or use of data, comprising the steps of: (a) storing in a memory store a structured, machine-readable representation of data that conforms to a machine-readable language; in which the data includes personal health or medical data; (b) automatically processing the structured representation of the data to analyse the personal health or medical data; in which the method includes the steps of (c) the machine-readable language representing a question in a memory in the structured, machine-readable representation of data; and (d) automatically generating a response to the question, using the following steps: (i) matching the question with the structured, machine-readable representations of data previously stored in the memory store; (ii) fetching and executing one or more computation units, wherein the computation units represent computational capabilities relevant to answering the question; (iii) fetching and execution of one or more reasoning

Abstract: A method of sampling output audio samples includes, during a packet loss concealment event, obtaining a sequence of previous output audio samples. At each time step during the event, the method includes generating a probability distribution over possible output audio samples for the time step. Each sample includes a respective probability indicating a likelihood that the corresponding sample represents a portion of an utterance at the time step. The method also includes determining a temperature sampling value based on a function of a number of time steps that precedes the time step, and an initial, a minimum, and a maximum temperature sampling value. The method also includes applying the temperature sampling value to the probability distribution to adjust a probability of selecting possible samples and randomly selecting one of the possible samples based on the adjusted probability. The method also includes generating synthesized speech using the randomly selected sample.

Abstract: Methods and apparatuses for conversation dialogue orchestration in virtual assistant communication sessions include a server that establishes a chat session between a virtual assistant (VA) application and a client device. The VA application captures an utterance generated by a user and processes the utterance to instantiate a dialogue behavior tree comprising workflow agents each associated with executable code for completing a corresponding workflow action. The VA application traverses the behavior tree to generate a response to the utterance, including evaluating one or more conditions associated with a workflow agent to determine whether to execute the code in the workflow agent, and when the conditions associated with the workflow agent are met, executing the code to complete the workflow action and storing a sub-response in a dialogue memory. The VA application coalesces the sub-responses to generate a final response and transmits the final response to the client device.

Abstract: A method of authoring a conversation service for a chatbot and a database includes receiving, from a user, a selection of a database, and connecting an authoring service of the chatbot to a table in the database; outputting, from the authoring service to the user, a question requesting a description of a subject matter of the table; receiving the description of the subject matter of the table; outputting to the user a question requesting an identification of a key column of the table that contains values that represent the subject matter of the table; receiving the identification of the key column of the table; and translating, by a natural language query service, the description of the subject matter of the table and the key column of the table into the conversation service, wherein the conversation service includes SQL statements suitable for querying the database table.

Abstract: A system is provided for machine-learning based identification and filtering of electronic network communication. In particular, the system may continuously monitor and pull electronic communications data from one or more networked computing systems in an enterprise environment. Based on the electronic communications data, the system may use machine learning algorithms to generate a database of associations between one or more users and one or more topics of interest. The system may then output one or more recommendations to one or more users for transmitting communications associated with the one or more topics of interest. In this way, the system may improve the efficiency of communications received and transmitted within a network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for end to end speech conversion are disclosed. In one aspect, a method includes the actions of receiving first audio data of a first utterance of one or more first terms spoken by a user. The actions further include providing the first audio data as an input to a model that is configured to receive first given audio data in a first voice and output second given audio data in a synthesized voice without performing speech recognition on the first given audio data. The actions further include receiving second audio data of a second utterance of the one or more first terms spoken in the synthesized voice. The actions further include providing, for output, the second audio data of the second utterance of the one or more first terms spoken in the synthesized voice.

Abstract: Methods and apparatus for bit prediction and data compression are provided. In the bit prediction method and apparatus, a specific Suffix Trie is used to maintain a statistical model to predict the next bit in a bit-stream. The statistic model provides the probability for each next or following bit in the bit stream, where an entropy encoder/decoder further encodes/decodes the predicted bit. The algorithm simplicity and its high performance relies on the combination of the specific method of Suffix Trie construction and growing, and the specific way to compute, update and propagate the probability across Suffix Trie nodes. The specific and particular method to grow the Suffix Trie in conjunction with the specific method to compute, update, and propagate the probability across Suffix Trie nodes are key aspects and subject matter of the present invention.

Abstract: An information processing method in which a computer executes processing includes: acquiring a sentence; specifying a word that appears immediately before or immediately after a first word in the acquired sentence by using a prediction model that predicts a word that appears immediately before or immediately after an input word; determining whether or not an estimated relationship between the first word and the second word in the sentence is appropriate on the basis of the specified word and a rule regarding a unit that corresponds to a relationship between words stored in a storage; and outputting information regarding the estimated relationship between the first word and the second word in a case where it is determined that the relationship is appropriate.

Abstract: A statistical language model may be used to simulate one or more users of a conversation. The statistical language model may be used to train a user to participate in a particular types of conversation by simulating communications by another type of user in the conversation. The communications may be simulated by selecting a simulation context from available simulation contexts and the simulation context may correspond to a difficulty level. Upon receiving a communication from a user, a responsive simulated communication may be generated by processing the received communication and the simulation context with the statistical language model. Upon completion of the simulation, another simulation context may be selected for the next simulation.

Abstract: Dependency trees are generated for questions and answers of a question answering (QA) corpus in which the answers are associated with the questions. Generating the dependency trees includes identifying root nodes. A word near an identified root node of one of the questions is compared to words of answers associated with the one of the questions. The word is determined to be in less than a threshold number of the associated answers. The word is identified as a stop word.

Abstract: A method includes receiving training data that includes unspoken text utterances, un-transcribed non-synthetic speech utterances, and transcribed non-synthetic speech utterances. Each unspoken text utterance is not paired with any corresponding spoken utterance of non-synthetic speech. Each un-transcribed non-synthetic speech utterance is not paired with a corresponding transcription. Each transcribed non-synthetic speech utterance is paired with a corresponding transcription. The method also includes generating a corresponding synthetic speech representation for each unspoken textual utterance of the received training data using a text-to-speech model. The method also includes pre-training an audio encoder on the synthetic speech representations generated for the unspoken textual utterances, the un-transcribed non-synthetic speech utterances, and the transcribed non-synthetic speech utterances to teach the audio encoder to jointly learn shared speech and text representations.

Abstract: A method for generating speech through multi-speaker neural text-to-speech (TTS) synthesis is provided. A text input may be received (1410). Speaker latent space information of a target speaker may be provided through at least one speaker model (1420). At least one acoustic feature may be predicted through an acoustic feature predictor based on the text input and the speaker latent space information (1430). A speech waveform corresponding to the text input may be generated through a neural vocoder based on the at least one acoustic feature and the speaker latent space information (1440).

Abstract: Techniques for predicting an answer to a question using a machine learning model are described. In some examples, the model predicts one or more answers to the question by: predicting at least two answers to the question using a first component of the question-answer model from a set of passages, generating, using a second component of the question-answer model, at least one question for each of the predicted at least two answers, and performing roundtrip predictions until each generated question only has one answer.

Abstract: The present disclosure relates to a method for searching a graph representing content of digital objects. A set of operations for traversing the graph may be determined according to a search request. The set of operations may be executed, resulting in intermediate result vectors of nodes and a result vector of nodes, wherein the result vector of nodes is associated with a result set of one or more object units of the digital objects. Intermediate result vectors may be selected from of the intermediate result vectors. A set of result entities may be identified. The set of result entities are entities which are part of the object units and part of entities represented by nodes of said selected intermediate result vectors. The set of result entities and the result set of object units may be provided as a result of the search request.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing speech separation. One of the methods includes obtaining a recording comprising speech from a plurality of speakers; processing the recording using a speaker neural network having speaker parameter values and configured to process the recording in accordance with the speaker parameter values to generate a plurality of per-recording speaker representations, each speaker representation representing features of a respective identified speaker in the recording; and processing the per-recording speaker representations and the recording using a separation neural network having separation parameter values and configured to process the recording and the speaker representations in accordance with the separation parameter values to generate, for each speaker representation, a respective predicted isolated audio signal that corresponds to speech of one of the speakers in the recording.

Abstract: Embodiments described herein are generally directed to training of classification models and their use by a chatbot to identify a customer-specified product support issue and provide appropriate troubleshooting guidance. According to an example, text describing an issue associated with a product line of a vendor is received via the chatbot. A vector representation of the issue is created using a word association model corresponding to the product line and trained based on a set of historical support cases relating to multiple supported issue categories for the product line. It is determined whether the issue matches a category within the supported issue categories for the product line by applying a classification model to the vector representation. When the determination is affirmative, an automated, interactive, conversational troubleshooting dialog is initiated with the user via the chatbot and guided based on a decision tree for the category within the product line.

Abstract: A method, computer system, and a computer program product for automatically generated question suggestions is provided. The present invention may include generating an intent-entity sequence representative of a current chat conversation with a user. The present invention may also include predicting, using a trained intent-entity prediction model, a next intent-entity sequence based on the generated intent-entity sequence representative of the current chat conversation. The present invention may further include converting the predicted next intent-entity sequence into at least one likely question suggestion to return back to the user.

Abstract: Systems and processes for operating an intelligent automated assistant are provided. In one example process, discourse input representing a user request is received. The process determines whether the discourse input relates to a device of an established location. In response to determining that the discourse input relates to a device of an established location, a data structure representing a set of devices of the established location is retrieved. The process determines, using the data structure, a user intent corresponding to the discourse input, the user intent associated with an action to be performed by a device of the set of devices, and a criterion to be satisfied prior to performing the action. The action and the device are stored in association with the criterion, where, in accordance with a determination that the criterion is satisfied, the action is performed by the device.