Abstract: Techniques for selectively requesting feedback from users are described. A system may receive a user input and perform an action responsive to the user input. The system may determine whether feedback should be requested from the user. Such determination may be based on various signals such as the user's present emotional state, whether the user input included profanity, whether the user input interrupted an output of the system, etc. When the system determines feedback should be requested, the system may select a feedback prompt pre-established by a domain, skill, or the like. After the system outputs the response to the user input, the system may output the feedback prompt.

Abstract: A method includes obtaining, using at least one processor of an electronic device, a base model trained to perform natural language understanding. The method also includes generating, using the at least one processor, a first model expansion based on knowledge from the base model. The method further includes training, using the at least one processor, the first model expansion based on first utterances without modifying parameters of the base model. The method also includes receiving, using the at least one processor, an additional utterance from a user. In addition, the method includes determining, using the at least one processor, a meaning of the additional utterance using the base model and the first model expansion.

Abstract: Techniques are described for intelligently identifying concept labels for a set of multiple documents where the identified concept labels are representative of and semantically relevant to the information contained by the set of documents. The technique includes extracting semantic units (e.g., paragraphs) from the set of documents and determining concept labels applicable to the semantic units based on relevance scores computed for the concept labels. The technique includes determining an initial set of concept labels for the set of documents based on the applicable concept labels. The technique further includes obtaining a reference hierarchy associated with the reference set of concept labels and determining a final set of concept labels for the set of documents using a reference hierarchy, the initial set of concept labels, and the relevance scores. The technique includes outputting information identifying the final set of concept labels for the set of documents.

Abstract: Methods and systems for performing a language processing task include setting an angular coordinate for a vector representation of each of a set of words, based on similarity of the words to one another. A radial coordinate is set for the vector representation of each word, according to hierarchical relationships between the words. A language processing task is performed based on hierarchical word relationships using the vector representations of the words.

Abstract: A system and method for providing aural guidance in imaging a document is disclosed. Visually impaired people may have difficulty when imaging a document with a mobile computing device. The system and method may analyze the image of the document in the imaging device, and output one or more aural directions in order for the user to move the document and/or camera for proper imaging of the document. In that regard, those with visual impairments may be assisted in imaging documents using a mobile computing device.

Abstract: Techniques for a natural language processing (NLP) system to implement more than one assistant are described. The NLP system may receive a natural language input from a device. The NLP system may also receive one or more signals representing one or more assistants to be implemented with respect to the natural language input. The NLP system may intelligently select an assistant to be invoked with respect to the natural language input. Once the assistant is selected, the NLP system may cause content, output to a user, to have characteristics specific to the assistant.

Abstract: Embodiments relate to a system, program product, and method for dictionary membership management directed at identifying ambiguity in semantic resources. A dictionary of seed terms is applied to a text corpus and matching items in the corpus are identified. The linguistic properties for each matching item are characterized and a context pattern of each matching item is constructed. Each context pattern is applied to the dictionary and matching content between the seed terms and the context pattern is identified and quantified. Lexicon items from the dictionary that have anomalous behavior reflected in the quantification are identified. One or more seed words identified as having anomalous behavior are selectively removed from the dictionary.

Abstract: During text-to-speech processing, a speech model creates synthesized speech that corresponds to input data. The speech model may include an encoder for encoding the input data into a context vector and a decoder for decoding the context vector into spectrogram data. The speech model may further include a voice decoder that receives vocal characteristic data representing a desired vocal characteristic of synthesized speech. The voice decoder may process the vocal characteristic data to determine configuration data, such as weights, for use by the speech decoder.

Abstract: Techniques and systems are described for content interest from interaction information. Keywords are extracted from digital content, and relevance values are determined based on the keywords that captures both the statistical and semantic significance of topics in the digital content through use of a network representation. Interest values for an entity are determined based on the relevance values and an interaction dataset, which capture both the statistical and semantic significance of the topics with respect to the entity. The interest values may be utilized to control output of digital content to a client device.

Abstract: Techniques are disclosed for processing natural language communications. A computing device receives a set of natural language communications. The computing device generates a word index from the communications and generates, from the word index, a set of topics, each topic including two or more words. For each topic, the computing device generates a score indicative of an amount of semantic information represented by the topic. The computing device then discards topics that are supersets or subsets of other topics. The computing device presents the remaining topics based on to the score of each topic.

Abstract: A content clarification server receives at least one language element entered by a user into a client computer, where the user works in a first area of specialization. The content clarification server extracts a set of concepts found in the at least one language element, and launches an auction bidding process for replacing original language in the at least one language element to content clarification providers who provide replacement language that clarifies a meaning of the at least one language element. The content clarification server filters out replacement language from content clarification providers that work in a second area of specialization that is different from the first area of specialization in which the user works, and identifies winning replacement language, from the filtered out replacement language, for the original language from one of the content clarification providers. The content clarification server replaces the original language with the winning replacement language.

Abstract: A voice controlled room automation system that includes a speaker device situated in a guest room, a hotel automation controller operatively coupled to one or more components in the guest room, and a web service operatively coupled to the speaker device and the hotel automation controller. The web service is configured to receiving voice commands from the speaker device, process the voice command using speech recognition, interpret the voice command to determine a corresponding command for the hotel automation controller, and transmit the corresponding command to the hotel automation controller. The hotel automation controller is configured to receive the corresponding command from the web service, and to carry out the corresponding command by interacting with one or more of the components in the room. In some cases, the hotel automation controller may be configured to initiate an announcement or query on the speaker device via the web service.

Abstract: A technique for intelligently identifying concept labels for a text fragment where the identified concept labels are representative of and semantically relevant to the information contained by the text fragment is provided. The technique includes determining, using a knowledge base storing information for a reference set of concept labels, a first subset of concept labels that are relevant to the information contained by the text fragment. The technique includes ordering the first subset of concept labels according to their relevance scores and performing dependency analysis on the ordered list of concept labels. Based on the dependency analysis, the technique includes identifying concept labels for a text fragment that are more independent (e.g., more distinct and non-overlapping) of each other, representative of and semantically relevant to the information represented by the text fragment.

Abstract: A system for short text identification can determine a plurality of topics and a representative noun that identifies each of the topics in a data repository. The system can determine a co-occurrence matrix for the training words stored in the corpus and determine a word vector embedding for each of the training words in the corpus to relate each of the training words in the corpus to other ones of the training words in the corpus in an n-dimensional vector space. The system can determine word tokens for words in short text in documents in the data repository that is separate and distinct from the corpus and determine sentence vectors for short text based on the word vectors in each short text and determine a plurality of topics in the documents based on clustering of sentence vectors, wherein the plurality of topics indicates topics that are predominant in the documents in the data repository.

Abstract: Systems and methods for sensing emotion in voice signals and dynamically changing suggestions in a call center are disclosed. According to one embodiment, a computer-implemented method comprises receiving a call from a customer at a call center. A sample of the call is recorded and the Mel-Frequency Cepstral coefficient is extracted from the sample. A machine learning model predicts an emotion of the customer and generates a confidence score for the emotion. A script of a call center agent is modified based on the emotion and the confidence score.

Abstract: An information processing device includes an input device, an output device, a processing device, a storage device, and a generating unit. The generating unit includes a first function of inputting text data; a second function of inputting a synonym dictionary including synonyms; a third function of extracting a first piece of text data from the multiple pieces of text data; a fourth function of extracting from the synonym dictionary synonyms in which a certain word in the first piece of text data is included; a fifth function of extracting from the text data second multiple pieces of text data including a synonym belonging to the extracted synonyms; and a sixth function of generating a score to evaluate a relationship between the first piece of text data and the extracted synonyms from a relationship between the first piece of text data and the second multiple pieces of text data.

Abstract: A method is described that estimates noise in an audio signal. An energy value for the audio signal is estimated and converted into the logarithmic domain. A noise level for the audio signal is estimated based on the converted energy value.

Abstract: A method for detecting voice, an apparatus for detecting voice, and a chip for processing voice are disclosed. The apparatus includes: a sub-band generation module and a voice activity detection module; wherein the sub-band generation module is configured to process a current time-domain signal frame to obtain sub-band time-domain signals, and the voice activity detection module is configured to determine, according to amplitudes of the sub-band time-domain signals in the current time-domain signal frame, whether the current time-domain signal frame is an effective voice signal. The apparatus for detecting voice may be practiced in a time domain, such that complexity of algorithms is lowered, and power consumption is reduced.

Abstract: According to one embodiment, a signal generation device includes one or more processors. The processors convert an acoustic signal and output amplitude and phase at a plurality of frequencies. The processors, for each of a plurality of nodes of a hidden layer included in a neural network that treats the amplitude and the phase as input, obtain frequency based on a plurality of weights used in arithmetic operation of the node. The processors generate an acoustic signal based on the plurality of obtained frequencies and based on amplitude and phase corresponding to each of the plurality of nodes.

Abstract: Methods, an encoder and a decoder are configured for transition between frames with different internal sampling rates. Linear predictive (LP) filter parameters are converted from a sampling rate S1 to a sampling rate S2. A power spectrum of a LP synthesis filter is computed, at the sampling rate S1, using the LP filter parameters. The power spectrum of the LP synthesis filter is modified to convert it from the sampling rate S1 to the sampling rate S2. The modified power spectrum of the LP synthesis filter is inverse transformed to determine autocorrelations of the LP synthesis filter at the sampling rate S2. The autocorrelations are used to compute the LP filter parameters at the sampling rate S2.