Abstract: In one aspect, method useful for goal-oriented dialog automation comprising includes the step of receiving an input message. The method includes the step of implementing an entity tagging operation on the input message. The method includes the step of tagging the message context of the input message to generate a tagged message context. The method includes the step of implementing semantic frame extraction from the tagged message context. The method includes the step of implementing an entity interpretation on the extracted frame. The method includes the step of accessing a database to determine a business schedule and a client profile. The business schedule and the client profile are related to the input message. The method includes the step of implementing a retrieval engine. The retrieval engine obtains one or more response templates. The method includes the step of generating a ranked list of candidate templates from the output of the retrieval engine.

Abstract: A dialogue system comprising: an input for receiving input data relating to a speech or text signal originating from a user; an output for outputting speech or text information specified by a dialogue act; and a processor configured to: update a belief state, the belief state comprising information corresponding to one or more dialogue options, each dialogue option comprising a slot and a corresponding slot value, based on the input signal; determine a dialogue act, wherein a dialogue act is determined by applying one or more rules to world state information, the world state comprising information relating to the dialogue, wherein rules are applied in two or more ordered stages for each dialogue turn, wherein one of the stages is a first update stage, comprising applying one or more further rules controlling updating of the world state information based on the belief state information, and another of the stages is an act selection stage, comprising determining the dialogue act by applying the one or more rule

Abstract: The present disclosure provides a computer system, method, and computer-readable medium for a computer processor to determine whether to report a consumer message regarding a vehicle to a regulatory agency. The computer system receives a message from a consumer. The computer system applies the message to a machine learning classifier trained to determine whether a category of vehicle system is described in the message, the machine learning classifier trained using category definitions and corresponding parts and symptoms labeled with the category definitions. The computer system determines whether the message includes a complaint based on an ontology defining a vehicle problem lexicon, a car part lexicon, and lexical patterns. The computer system extracts MVS terms from the message. The computer system determines to report the message if the message includes a complaint related to at least one category of vehicle system and includes a set of the MVS terms.

Abstract: A system, method and computer product for extracting an activity from recordings. The method comprises searching for signals representing plural versions of a track, determining feature representations of the plural versions of the track identified in the searching, aligning the feature representations determined in the determining, and extracting a time varying activity signal from the feature representations aligned in the aligning.

Abstract: A method, system and computer-usable medium for classifying a source document using sub-documents identified in the source document. The method, system, and computer-usable medium are used to access the source document from electronic memory. The source document is electronically searched to detect markers indicative of whether the source document includes one or more sub-documents. Incongruities in the source document are located using the detected markers and the source document is split into sub-documents at the located incongruities. Each of the sub-documents is classified. The sub-documents are joined as a re-assembled source document with classifications including classifications for one or more of the sub-documents.

Abstract: A computing device includes a communication interface configured to interface and communicate with a communication system, an audio interface configured to interface and communicate with a user, a memory that stores operational instructions, and processing circuitry operably coupled to the communication interface, the audio interface, and to the memory that is configured to execute the operational instructions to perform various operations. The computing device monitors audio content, maintains a running buffer of most recent audio content, and detects a wake word command of the user. When detected, the computing device processes the most recent audio content including the wake word command of the user to determine validity/invalidity whether the wake word command of the user is invalid based on the most recent audio content. When invalid, the computing device rejects the wake word command of the user and continues to monitor the audio content and maintain the running buffer.

Abstract: Techniques for a recursive deep-learning approach for performing speech synthesis using a repeatable structure that splits an input tensor into a left half and right half similar to the operation of the Fast Fourier Transform, performs a 1-D convolution on each respective half, performs a summation and then applies a post-processing function. The repeatable structure may be utilized in a series configuration to operate as a vocoder or perform other speech processing functions.

Abstract: A method and an apparatus for estimating speech signal in split-domain is disclosed. The method includes performing LP analysis on a noisy speech signal to generate a first plurality of LPC and a first residual signal. The method also includes estimating speech LPC spectrum to generate cleaned LPC. The method further includes estimating speech residual spectrum to generate cleaned residual signal. The method also includes synthesizing output signals based on the cleaned LPC and the cleaned residual signal.

Abstract: A device capable of splitting user input into phrases is presented. The disclosed device leverages multiple phrase splitting models to generate one or more possible split locations. The possible split locations can be derived based on leveraging multiple phrase splitting models. Each model contributes its suggested split locations to the set of possible split locations according to an implementation of a phrase splitting kernel algorithm that weights each model's suggestions.

Abstract: Embodiments are directed to a system, computer program product, and method for dynamic facet dictionary management. As one or more annotations are applied to a document collection, electronic text and associated facets are identified. Additional facets and facet values are identified and selectively applied to a knowledge base. A dictionary comprised of facets and associated facet values is constructed from the selective application. Application of the dictionary to the knowledge base identifies and returns a targeted document collection. Accordingly, facet mining and dictionary construction are dynamically applied to the knowledge base.

Abstract: Disclosed herein are systems, methods, and non-transitory computer-readable storage media for generating domain-specific speech recognition models for a domain of interest by combining and tuning existing speech recognition models when a speech recognizer does not have access to a speech recognition model for that domain of interest and when available domain-specific data is below a minimum desired threshold to create a new domain-specific speech recognition model. A system configured to practice the method identifies a speech recognition domain and combines a set of speech recognition models, each speech recognition model of the set of speech recognition models being from a respective speech recognition domain. The system receives an amount of data specific to the speech recognition domain, wherein the amount of data is less than a minimum threshold to create a new domain-specific model, and tunes the combined speech recognition model for the speech recognition domain based on the data.

Abstract: Aspects of the subject technology relate to a method for using a voice command for multiple computing devices. First voice input data is received from a first computing device associated with a user account, where the first voice input data comprises a first voice command captured at the first computing device. Second voice input data is received from a second computing device associated with the user account where the second voice input data comprises a second voice command captured at the second computing device. An intended voice command is determined based on the obtained first and second voice input data. Based on the intended voice command, a first target computing device is determined. First instructions associated with the intended voice command are provided to the first target computing device for execution.

Abstract: A sentence generating apparatus includes an encoder configured to generate a first sentence embedding vector by applying trained result data to a first paraphrased sentence of an input sentence, an extractor configured to extract verification sentences in a preset range from the generated first sentence embedding vector, and a determiner configured to determine a similarity of the first paraphrased sentence to the input sentence based on comparing the verification sentences to the input sentence.

Abstract: Methods and apparatuses are described for automated computer text classification and routing using artificial intelligence transfer learning. A server trains a word embedding model using one-hot vectors of word pairs from a filtered first corpus of unstructured computer text and a filtered second corpus of unstructured computer text, using an artificial intelligence neural network. The server trains a long short-term memory model using vector matrices that correspond to sentences in the filtered second corpus of unstructured computer text, and labels. The server receives a message, generates a matrix for each sentence in the message by applying the trained word embedding model, generates one or more labels, and a probability for each label, for each sentence in the message by applying the trained long short-term memory model, and routes the message to a second client computing device based upon an assigned label.

Abstract: Provided is an electronic device that includes a first processor for receiving an audio signal, performing first voice recognition on the audio signal, and transferring a driving signal to a second processor based on a result of the first voice recognition. The second processor performs second voice recognition based on a voice signal by the first voice recognition or the audio signal, in response to the driving signal.

Abstract: Systems and methods are disclosed for displaying electronic multimedia content to a user. One computer-implemented method for manipulating electronic multimedia content includes generating, using a processor, a speech model and at least one speaker model of an individual speaker. The method further includes receiving electronic media content over a network; extracting an audio track from the electronic media content; and detecting speech segments within the electronic media content based on the speech model. The method further includes detecting a speaker segment within the electronic media content and calculating a probability of the detected speaker segment involving the individual speaker based on the at least one speaker model.

Abstract: Embodiments of the present disclosure provide a language recognition method and apparatus, a device and a computer storage medium. In an aspect, in the embodiments of the present disclosure, after the Nth speech segment included by the speech signal is received, language recognition is performed according to already-received previous N speech segments to obtain the score of each language in at least one language, N being 2, 3, 4 . . . ; therefore, if there exists a langue whose score reaches the designated threshold, the language whose score reaches the designated threshold is considered as the language matched with the speech signal. Therefore, the technical solutions according to embodiments of the present disclosure solve the problem in the prior art that the efficiency of language recognition is lower so that the language recognition cannot be applied to an application scenario in which the recognition result needs to be obtained quickly.

Abstract: Methods, systems, and computer programs are presented for training, with adversarial constraints, a student model for speech recognition based on a teacher model. One method includes operations for training a teacher model based on teacher speech data, initializing a student model with parameters obtained from the teacher model, and training the student model with adversarial teacher-student learning based on the teacher speech data and student speech data. Training the student model with adversarial teacher-student learning further includes minimizing a teacher-student loss that measures a divergence of outputs between the teacher model and the student model; minimizing a classifier condition loss with respect to parameters of a condition classifier; and maximizing the classifier condition loss with respect to parameters of a feature extractor. The classifier condition loss measures errors caused by acoustic condition classification. Further, speech is recognized with the trained student model.

Abstract: A method and system are disclosed for analyzing text affinity among a plurality of social media communications, comprising dividing a first social media communication into first plurality of social media communication threads; dividing a second social media communication into a second plurality of social media communication threads; performing a text affinity analysis operation between respective threads of the first plurality of social media communication threads and the second plurality of social media communication threads; and, determining a level of intervention to perform based upon the text affinity analysis operation.

Abstract: A system configured to perform power normalization for voice enhancement. The system may identify active intervals corresponding to voice activity and may selectively amplify the active intervals in order to generate output audio data at a near uniform loudness. The system may determine a variable gain for each of the active intervals based on a desired output loudness and a flatness value, which indicates how much a signal envelope is to be modified. For example, a low flatness value corresponds to no modification, with peak active interval values corresponding to the desired output loudness and lower active intervals being lower than the desired output loudness. In contrast, a high flatness value corresponds to extensive modification, with peak active interval values and lower active interval values both corresponding to the desired output loudness. Thus, individual words may share the same peak power level.