Abstract: The autonomous avatar driver is useful in association with language sources. A sourcer may receive dialog from the language source. It may also, in some embodiments, receive external data from data sources. A segmentor may convert characters, represent particles and split dialog. A parser may then apply a link grammar, analyze grammatical mood, tag the dialog and prune dialog variants. A semantic engine may lookup token frames, generate semantic lexicons and semantic networks, and resolve ambiguous co-references. An analytics engine may filter common words from dialog, analyze N-grams, count lemmatized words, and analyze nodes. A pragmatics analyzer may resolve slang, generate knowledge templates, group proper nouns and estimate affect of dialog. A recommender may generate tag clouds, cluster the language sources into neighborhoods, recommend social networking to individuals and businesses, and generate contextual advertising.

Abstract: Techniques for providing virtual assistants to assist users during a voice communication between the users. For instance, a first user operating a device may establish a voice communication with respective devices of one or more additional users, such as with a device of a second user. For instance, the first user may utilize her device to place a telephone call to the device of the second user. A virtual assistant may also join the call and, upon invocation by a user on the call, may identify voice commands from the call and may perform corresponding tasks for the users in response.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for neural translation systems with rare word processing. One of the methods is a method training a neural network translation system to track the source in source sentences of unknown words in target sentences, in a source language and a target language, respectively and includes deriving alignment data from a parallel corpus, the alignment data identifying, in each pair of source and target language sentences in the parallel corpus, aligned source and target words; annotating the sentences in the parallel corpus according to the alignment data and a rare word model to generate a training dataset of paired source and target language sentences; and training a neural network translation model on the training dataset.

Abstract: Methods and apparatus to audio watermarking and watermark detection and extracted are described herein. According to an example method, an identifier is encoded in media content when a different identifier has been previously encoded. According to another example method, messages decoded from media content are validated to provide improved decoding accuracy. In another example method, decoded symbols are stored in memory and synchronization symbols are located to detect a message encoded in media content.

Abstract: Contextual note taking is described. A note taking assistant can receive an indication of a specific presentation session. This indication can be used by the note taking assistant to access information or content related to the session. The note taking assistant can receive specific presentation session content, which includes identifiable context images. Identifiable context images are meant to define an individual page, an individual slide, or other atomic unit in the presentation. The note taking assistant operates by receiving a navigation message, changing the current assistant context image to a current presenter context image based on the navigation message; receiving a speech-to-text message comprising a unit of text; displaying the current presenter context image, and displaying the unit of text associated with the current presenter context image; and storing the unit of text associated with the current presenter context image.

Abstract: A gain adjustment apparatus (60) for use in decoding of audio that has been encoded with separate gain and shape representations includes an accuracy meter (62) configured to estimate an accuracy measure (A(b)) of the shape representation (Ñ(b)), and to determine a gain correction (gc(b)) based on the estimated accuracy measure (A(b)). It also includes an envelope adjuster (64) configured to adjust the gain representation (Ê(b)) based on the determined gain correction.

Abstract: A noise suppression device includes an estimating unit that estimates, from a feature quantity representing the feature in each frequency range of a first acoustic signal which represents sound, the noise component of the feature quantity; a calculating unit that calculates, from the feature quantity and the noise component for each frequency range, a first suppression coefficient to be used in suppressing noise included in the first acoustic signal; a first attenuating unit that attenuates the first suppression coefficient in the time domain and calculates a second suppression coefficient; a second attenuating unit that attenuates the second suppression coefficient in the frequency domain and calculates a third suppression coefficient; and a generating unit that estimates, from the feature quantity and the third suppression coefficient, a voice component of the feature quantity and generates a second acoustic signal in which the noise included in the first acoustic signal is suppressed.

Abstract: Features are disclosed for maintaining data that can be used to personalize spoken language understanding models, such as speech recognition or natural language understanding models. The personalization data can be used to update the models based on some or all of the data. The data may be obtained from various data sources, such as applications or services used by the user. Personalized spoken language understanding models may be generated or updated based on updates to the personalization data or some other portion of the stored personalization data. Generation of personalized spoken language understanding models may be prioritized such that the generation process accommodates multiple users.

Abstract: Incremental speech recognition results are generated and used to determine a user's intent from an utterance. Utterance audio data may be partitioned into multiple portions, and incremental speech recognition results may be generated from one or more of the portions. A natural language understanding module or some other language processing module can generate semantic representations of the utterance from the incremental speech recognition results. Stability of the determined intent may be determined over the course of time, and actions may be taken in response to meeting certain stability thresholds.

Abstract: A computer-implemented method includes identifying a first set of utterances from a plurality of utterances. The plurality of utterances is associated with a conversation and transmitted via a plurality of audio signals. The computer-implemented method further includes mining the first set of utterances for a first context. The computer-implemented method further includes determining that the first context associated with the first set of utterances is not relevant to a second context associated with the conversation. The computer-implemented method further includes dynamically muting, for at least a first period of time, a first audio signal in the plurality of audio signals corresponding to the first set of utterances. A corresponding computer system and computer program product are also disclosed.

Abstract: Mechanisms, in a natural language processing (NLP) system are provided. The NLP system receives a plurality of communications associated with a communication system, over a predetermined time period, from a plurality of end user devices. The NLP system identifies, for each communication in the plurality of communications, a user submitting the communication to thereby generate a set of users comprising a plurality of users associated with the plurality of communications. The NLP system retrieves a user model for each user in the set of users, which specifies at least one attribute of a corresponding user. The NLP system generates an aggregate user model that aggregates the at least one attribute of each user in the set of users together to generate an aggregate representation of the attributes of the plurality of users in the set of users. The NLP system performs a cognitive operation based on the aggregate user model.

Abstract: Key phrase detection techniques for applications such as wake on voice are discussed include performing a vectorized operation on a multiple element acoustic score vector for a current time instance including a single state rejection model score and scores for a multiple state key phrase model and a multiple element state score vector for a previous time instance including a previous state score for the single state rejection model and previous state scores for the multiple state key phrase model to generate a multiple element score summation vector and a second vectorized operation on the multiple element score summation vector to determine a multiple element state score vector for the current time instance. The multiple element state score vector for the current time instance may then be evaluated to determine whether received audio input includes a key phrase corresponding to the multiple state key phrase model.

Abstract: One example discloses an apparatus for sound signal detection, comprising: a first wireless device including a first pressure sensor having a first acoustical profile and configured to capture a first set of acoustic energy within a time window; wherein the first wireless device includes a wireless signal input; wherein the first wireless device includes a processing element configured to: receive, through the wireless signal input, a second set of acoustic energy captured by a second pressure sensor, having a second acoustical profile, within a second wireless device and within the time window; apply a signal enhancement technique to the first and second sets of acoustic energy based on the first and second acoustical profiles; search for a predefined sound signal within the enhanced sets of acoustic energy; and initiate a subsequent set of sound signal detection actions if the search finds the sound signal.

Abstract: A speech recognition system uses, in one embodiment, an extended phonetic dictionary that is obtained by processing words in a user's set of databases, such as a user's contacts database, with a set of pronunciation guessers. The speech recognition system can use a conventional phonetic dictionary and the extended phonetic dictionary to recognize speech inputs that are user requests to use the contacts database, for example, to make a phone call, etc. The extended phonetic dictionary can be updated in response to changes in the contacts database, and the set of pronunciation guessers can include pronunciation guessers for a plurality of locales, each locale having its own pronunciation guesser.

Abstract: Automatic semantic analysis for characterizing and correlating literary elements within a digital work of literature is accomplished by employing natural language processing and deep semantic analysis of text to create annotations for the literary elements found in a segment or in the entirety of the literature, a weight to each literary element and its associated annotations, wherein the weight indicates an importance or relevance of a literary element to at least the segment of the work of literature; correlating and matching the literary elements to each other to establish one or more interrelationships; and producing an overall weight for the correlated matches.

Abstract: Automatic semantic analysis for characterizing and correlating literary elements within a digital work of literature is accomplished by employing natural language processing and deep semantic analysis of text to create annotations for the literary elements found in a segment or in the entirety of the literature, a weight to each literary element and its associated annotations, wherein the weight indicates an importance or relevance of a literary element to at least the segment of the work of literature; correlating and matching the literary elements to each other to establish one or more interrelationships; and producing an overall weight for the correlated matches.

Abstract: A method of voice recognition and an electronic apparatus are described with the method of voice recognition being applied in an electronic apparatus. The method includes taking i=1 and detecting corresponding i-th voice sub-information at a moment Ti when the electronic apparatus detects that a user starts to talk at a moment T0, wherein the i-th voice sub-information voice information from the moment T0 to the moment Ti, the i-th voice sub-information is partial voice information of voice information with integral semantic corresponding to a moment Tj after the moment T0 to the moment Ti, and i is an integer greater than or equal to 1; and analyzing the i-th voice sub-information to obtain M results of analysis, M being an integer greater than or equal to 1.

Abstract: A computer-implemented method for analyzing content written on a board, on which a text and/or a drawing are made, is disclosed. The method includes obtaining content data including a series of images, which captures content being written on the board. The method also includes obtaining utterance data representing a series of utterances, which is associated with the series of the images. The method further includes extracting a section from the series of the utterances based on a change in topics and recognizing a content block for the section from the content data. The content block includes one or more content parts written during the section. The method includes further calculating evaluation value for the content block by using one or more utterances included in the section.

Abstract: A human-machine dialog system including a first part including a data processing unit and an electrical power supply unit, a second part including a number of human-machine dialog members each provided with a human-machine dialog interface, and a fastening support arranged to accommodate each human-machine dialog member in a removable and interchangeable manner, and a communication and power supply bus linking the power supply unit and the processing unit of the first part to each human-machine dialog member.

Abstract: Provided is a method of managing a server having stored thereon a phrase table containing phrase pairs each being a pair of a phrase in a first language and a phrase in a second language different from the first language. In the method, a processor receives combinations of the phrase pairs used to generate respective candidate target texts which are generated by translation of a source text in the first language into a target text in the second language as candidates for the target text in the second language, and records combination information (for example, a data structure and paths) in the server, the combination information identifying, in the phrase pairs contained in the phrase table, the combinations of the phrase pairs used to generate the respective candidate target texts.