Abstract: One embodiment provides a method, including receiving, at an audio capture device, a customized activation cue; identifying, using a processor, contextual information associated with a user; analyzing, using the contextual information, characteristics of the customized activation cue; identifying, based on the analyzation, a uniqueness associated with the customized activation cue; and responsive to said identifying, notifying a user that the customized activation cue has inadequate uniqueness. Other aspects are described and claimed.

Abstract: A method includes obtaining an input text, identifying a first term in the input text, and accessing lexicon data to identify a first entry corresponding to the first term. The first entry includes core data corresponding to domain-independent lexical information for the first term, and non-core data corresponding to domain-specific lexical information for the first term. The method also includes determining that the non-core data of the first entry identifies a second term in the input text as a modifier of the first term. The method further includes generating a partially parsed and bracketed version of the input text. The partially parsed and bracketed version indicates that the second term modifies the first term in the input text. The method also includes generating a parsed version of the input text based on the partially parsed and bracketed version of the input text.

Abstract: A system and method are presented for outlier identification to remove poor alignments in speech synthesis. The quality of the output of a text-to-speech system directly depends on the accuracy of alignments of a speech utterance. The identification of mis-alignments and mis-pronunciations from automated alignments may be made based on fundamental frequency methods and group delay based outlier methods. The identification of these outliers allows for their removal, which improves the synthesis quality of the text-to-speech system.

Abstract: A method and system are provided. The method includes separating a predicate that specifies a set of events into a temporal part and a non-temporal part. The method further includes comparing the temporal part of the predicate against a predicate of a known window type. The method also includes determining whether the temporal part of the predicate matches the predicate of the known window type. The method additionally includes replacing (i) the non-temporal part of the predicate by a filter, and (ii) the temporal part of the predicate by an instance of the known window type, responsive to the temporal part of the temporal predicate matching the predicate of the known window type. The instance is parameterized with substitutions used to match the temporal part of the predicate to the predicate of the known window type.

Abstract: Systems, methods, and devices for outputting visual indications regarding voice-based interactions are described. A first speech-controlled device detects spoken audio corresponding to a voice message to a second speech-controlled device. The first device captures the audio and sends audio data corresponding to the captured audio to a server. The server performs speech processing on the audio data to determine a recipient and message content. The server then determines a second speech-controlled device associated with the recipient and sends the message content to the recipient's second speech-controlled device. Thereafter, the server receives an indication from the recipient's speech-controlled device that the second device is detecting speech, presumably in response to the original message. The server then causes a visual indication to be output by the first speech-controlled device, with the visual indication representing the recipient-speech controlled device is detecting speech.

Abstract: A computer program product including a computer-readable storage medium that has program instructions embodied therewith. The program instructions executable to cause the processor to perform operations including obtaining an input text at a text parser, identifying a first term in the input text, and accessing lexicon data to identify a first entry corresponding to the first term. The first entry includes non-core data that corresponds to domain-specific lexical information for the first term. The operations further include determining that the non-core data identifies a second term in the input text as a modifier of the first term. The operations also include generating a partially parsed and bracketed (PPB) version of the input text. The PPB version indicates that the second term modifies the first term in the input text. The operations further include generating a parsed version of the input text based on the PPB version of the input text.

Abstract: A system for creating a synthetic voice identifier may include a plurality of synthesized voice authorization (SVA) devices and a biometric combinatory device (BCD). The SVAs may be communicatively coupled to the BCD via a network and may communicate utilizing a markup language. The SVA devices may capture an audio signal of a voice of a user, modify the audio signal with a randomized audio frequency signal to generate a modified audio signal, and communicate, the modified audio signal as a synthesized voice signal associated with the user.

Abstract: A voice dialog device, comprises a sight line detection unit configured to detect a sight line of a user; a voice processing unit configured to obtain voice pronounced by the user and a result of recognizing the voice; a dialog determination unit configured to determine whether or not the voice dialog device has a dialog with the user; and an answer generation unit configured to generate an answer, based on a result of recognizing the voice, wherein the dialog determination unit determines whether or not the user has started the dialog, based on both the sight line of the user and the obtained voice.

Abstract: Mechanisms are provided to perform embedding of content of a natural language document. The mechanisms receive a document data object of an electronic document and analyze a structure of the electronic document to identify one or more structural document elements that have a relationship with the document data object. A dependency data structure is generated, representing the electronic document, where edges define relationships between document elements and at least one edge represents at least one relationship between the one or more structural document elements and the document data object. The mechanisms embed the document data object based on the at least one relationship to thereby represent the document data object as a vector data structure. The mechanisms perform natural language processing on the portion of natural language content based on the vector data structure. The one or more structural document elements are non-local non-contiguous with the document data object.

Abstract: A corpus pattern paraphrasing method, system, and non-transitory computer readable medium, include an analyzing circuit configured to analyze a corpus of sentences stored in a database to determine regular structures including a plurality of substitute words for verbs expressed as patterns and apply deep learning of the regular structures over the patterns, a representative word determining circuit configured to determine a plurality of representative words that represents each class of word of the regular structures, and an aligning circuit configured to align word slots of a paraphrase pattern of the classes of words replaced with substitute words and representative words in the paraphrase pattern to give a same semantic meaning to the paraphrase pattern as a sentence of the corpus of sentences.

Abstract: A mechanism is provided in a computing device configured with instructions executing on a processor of the computing device to implement a timeline generation system, for automated timeline completion. The timeline generation system executing on the processor of the computing device identifies a plurality of events in documents in a corpus of information. The timeline generation system places the plurality of events in a partial timeline data structure. The timeline generation system selects an event progression from an event progression knowledge base. The timeline generation system aligns the selected event progression to the partial timeline data structure. The timeline generation system identifies a set of events missing from the partial timeline data structure. The timeline generation system maps the set of events missing from the partial timeline data structure to the partial timeline based on the selected event progression to form a completed timeline data structure.

Abstract: A device, system, and method whereby a speech-driven system used in an industrial environment distinguishes speech obtained from users of the system from other background sounds. In one aspect, the present system and method provides for a first audio stream from a user microphone collocated with a source of human speech (that is, a user) and a second audio stream from a environmental microphone which is proximate to the source of human speech but more remote than the user microphone. The audio signals from the two microphones are asynchronous. A processor is configured to identify a common, distinctive sound event in the environment, such as an impulse sound or a periodic sound signal. Based on the common sound event, the processor provides for synchronization of the two audio signals.

Abstract: Provided is a method and computer program product for producing an enhanced audio signal for an output device from audio signals received by 2 or more microphones in close proximity to each other. For example, one embodiment of the present invention comprises the steps of receiving a first input audio signal from the first microphone, digitizing the first input audio signal to produce a first digitized audio input signal, receiving a second input audio input signal from the second microphone, digitizing the second input audio input signal to produce a second digitized audio input signal, using the first digitized audio input signal as a reference signal to an adaptive prediction filter, using the second digitized audio input signal as input to said adaptive prediction filter and finally adding a prediction result signal from the adaptive prediction filter to the first digitized audio input signal to produce the enhanced audio signal.

Abstract: Many processes for audio signal processing can benefit from voice activity detection, which aims to detect the presence of speech as opposed to silence or noise. The present disclosure describes, among other things, leveraging energy-based features of voice and insights on first and second formant frequencies of vowels to provide a low-complexity and low-power voice activity detector. A pair of two channels is provided whereby each channel is configured to detect voice activity in respective frequency bands of interest. Simultaneous activity detected in both channels can be a sufficient condition for determining that voice is present. More channels or pairs of channels can be used to detect different types of voices to improve detection and/or to detect voices present in different audio streams.

Abstract: Identification of an intent of a conversation can be useful for real-time or post-processing purposes. According to example embodiments, a method, and corresponding apparatus of identifying at least one intent-bearing utterance in a conversation, comprises determining at least one feature for each utterance among a subset of utterances of the conversation; classifying each utterance among the subset of utterances, using a classifier, as an intent classification or a non-intent classification based at least in part on a subset of the at least one determined feature; and selecting at least one utterance, with intent classification, as an intent-bearing utterance based at least in part on classification results by the classifier. Through identification of an intent bearing utterance, a call center for example, can provide improved service for callers through, for example, more effective directing of a call to a live agent.

Abstract: Methods, apparatus, systems, and computer-readable media are provided for storing incomplete dialog sessions between a user and an automated assistant in order that the dialog sessions can be completed in furtherance of certain actions. While interacting with an automated assistant, a user can become distracted and not complete the interaction to the point of the automated assistant performing some action. In response, the automated assistant can store the interaction as a dialog session. Subsequently, the user may express interest, directly or indirectly, in completing the dialog session, and the automated assistant can provide the user with a selectable element that, when selected, causes the dialog session to be reopened. The user can then continue the dialog session with the automated assistant in order that the originally intended action can be performed by the automated assistant.

Abstract: One or more processors deconstruct, by utilizing natural language processing (NLP), a text product review into multiple n-grams, where each of the multiple n-grams is a sequence of words from the text product review, and where each of the multiple n-grams is a review of a particular component from the multiple components of the product. The processor(s) generate a component numeric rating value (CNRV) for each of the multiple n-grams using machine learning techniques such as collaborative filtering, and store CNRVs for the multiple components of the product together with identifiers of particular components of the product. The processor(s) receive a request for a particular CNRV for the particular component of the product, retrieve the particular CNRV, and then transmit the particular CNRV to a requesting device of the request in order to adjust an operational feature of the requesting device.

Abstract: The present disclosure provides a voice recognition method, when receiving a voice signal, an image acquisition device is controlled to acquire images, and when the voice signal being over, the image acquisition device is controlled to stop acquiring the image; the received voice signal is recognized to obtain a voice signal recognition result; accuracies of the voice signal recognition result and the lip-reading recognition result is calculated, the recognition result with a higher accuracy is taken as a current voice recognition result. The present disclosure also provides a voice recognition system. The present disclosure improves the accuracy of the voice recognition.

Abstract: Disclosed herein are systems and methods for removing reverberation from signals. The systems and methods can be applicable to audio signals, for example, to voice, musical instrument sounds, and the like. Signals such as the vowel sounds in speech and the sustained portions of many musical instrument sounds can be composed of a fundamental frequency component and a series of harmonically related overtones. The systems and methods can exploit the intrinsically high degree of mutual correlation among the overtones. When such signals are passed through a reverberant channel, the degree of mutual correlation among the partials can be reduced. An inverse channel filter for the removal of reverberation can be found by employing an adaptive filter technique that maximizes the cross-correlation among signal overtones.

Abstract: Natural language system question classifier, semantic representations, and logical form template techniques and systems are described. In one or more implementations, a natural language input is classified as corresponding to respective ones of a plurality of classes of questions. A semantic intent of the natural language input is extracted as a semantic entity and a semantic representation. Question classification labels that classify the question included in the natural language input is then used to select at least one of a plurality of logical form templates. The semantic intent that is extracted from the natural language input is then used to fill in the selected logical form templates, such as to fill in entity, subject, predicate, and object slots using the semantic entity and semantic representation. The filled-in logical form template is then mapped to form a database query that is then executed to query a database to answer the question.