Abstract: A method for training a deep neural network (DNN), comprises receiving and formatting speech data for the training, performing Hessian-free sequence training (HFST) on a first subset of a plurality of subsets of the speech data, and iteratively performing the HFST on successive subsets of the plurality of subsets of the speech data, wherein iteratively performing the HFST comprises reusing information from at least one previous iteration.

Abstract: A system and method are described for engaging an audience in a conversational advertisement. A conversational advertising system converses with an audience using spoken words. The conversational advertising system uses a speech recognition application to convert an audience's spoken input into text and a text-to-speech application to transform text of a response to speech that is to be played to the audience. The conversational adverting system follows an advertisement script to guide the audience in a conversation.

Abstract: An audio device and a method thereof are provided. The method is adopted by an audio device to detect a voice, wherein the audio device is coupled to a host device. The method includes an acoustic conversion circuit converting an acoustic wave into an analog audio signal; an analog-to-digital converter (ADC) converting the analog audio signal into digital audio data; a first-level voice detection circuit detecting voice activity in the analog audio signal; a second-level voice detection circuit detecting a beginning syllable of a key phrase in the digital audio data when the voice activity is detected in the digital audio data; and a third-level voice detection circuit detecting the key phrase from the digital audio data only when the beginning syllable of the key phrase is detected in the digital audio data.

Abstract: A computer system executing a computer audio application such as video conferencing applies audio detection and speech recognition to an input audio stream to generate respective audio detection and speech recognition signals. A function is applied to the audio detection and speech recognition signals to generate a non-speech audio detection signal identifying presence of non-speech audio in the input audio stream when the audio detection signal is asserted and the speech recognition signal is not asserted. A control or indication action is performed in the computer system based on assertion of the non-speech audio detection signal.

Abstract: This illustrative embodiments provide a mechanism for correcting a phonetically sourced spelling mistake. The mechanism receives a language text string comprising at least one spelling mistake word and transcribes the at least one spelling mistake word into a phonetic form of the spelling mistake word using a phonetic dictionary. The mechanism locates a correctly spelled phonetic form from a phonetic form dictionary having shortest edit distance between characters of the correctly spelled phonetic form word and the phonetic transcription whereby the phonetic form dictionary comprises correctly spelled words and associated phonetic forms of the correctly spelled words. The mechanism substitutes the correctly spelled word for the spelling mistake word in the text string.

Abstract: Disclosed is a system, method, and computer program product for allowing an entity to access social media data, and to perform term analysis upon that data. The approach is capable of accessing data across multiple types of internet-based sources of social data and commentary. A user interface is provided that allows the user to view and interact with the results of performing term analysis.

Abstract: Embodiments relate to facilitating a meeting. A method for facilitating a meeting of a group of participants is provided. The method generates a graph of words from speeches of the participants as the words are received from the participants. The method partitions the group of participants into a plurality of subgroups of participants. The method performs a graphical text analysis on the graph to identify a cognitive state for each participant and a cognitive state for each subgroup of participants. The method informs at least one of the participants about the identified cognitive state of a participant or a subgroup of participants.

Abstract: A method for determining a voice command shortcut includes receiving a first voice command providing instructions for performing a particular task and a second voice command providing additional instructions for performing the same task. The voice command shortcut may be used in place of the first and second voice commands, which are typically submitted in response to system prompts. The availability of a voice command shortcut is determined based on the first and second voice commands. If a voice command shortcut is available, an audible and/or visual notification may be provided to inform the user of the available voice command shortcut.

Abstract: An approach is provided for ranking candidate answers to a natural language question. Candidate answers to a natural language question received from a mobile device are generated. First contextual information about a user of the mobile device is identified. A prioritization of definitions of terms is determined. Based on the prioritization, a lexicon of the terms is generated. Using mobile-based time series manipulation and pattern recognition and based on historical usage of the mobile device, a location of the user, an environment of the user, and a bodily function of the user, second contextual information is forecasted. Based on a word sense disambiguation of the terms in the lexicon and an adjustment of the prioritization, the candidate answers are modified and then ranked. The highest ranked candidate answer is more likely to be a correct answer to the natural language question than the other candidate answers.

Abstract: Embodiments relate to facilitating a meeting. A method for facilitating a meeting of a group of participants is provided. The method generates a graph of words from speeches of the participants as the words are received from the participants. The method partitions the group of participants into a plurality of subgroups of participants. The method performs a graphical text analysis on the graph to identify a cognitive state for each participant and a cognitive state for each subgroup of participants. The method informs at least one of the participants about the identified cognitive state of a participant or a subgroup of participants.

Abstract: The techniques disclosed herein allow a user to synchronize the playing and displaying of digital content on an electronic device. The device may render a first portion of digital content so it may be displayed. The device may also play a segment of the digital content as audio using text to speech software. The device may also render a second portion of digital content for display depending on whether the position of the last word read is greater than the last position in the first portion of digital content.

Abstract: Methods and apparatus for language translation in a computing environment associated with a virtual application are presented. For example, a method for providing language translation includes determining languages of a user and a correspondent; determining one or more sequences of translators; determining a selected sequence of selected translators from the one or more sequences of the translators; requesting a change in virtual locations, within the computing environment associated with the virtual application, of one or more selected translator virtual representations of the selected translators to a virtual meeting location within the computing environment associated with the virtual application; and changing virtual locations of the one or more selected translator virtual representations to the virtual meeting location.

Abstract: A method and system is disclosed for building a speech database for a text-to-speech (TTS) synthesis system from multiple speakers recorded under diverse conditions. For a plurality of utterances of a reference speaker, a set of reference-speaker vectors may be extracted, and for each of a plurality of utterances of a colloquial speaker, a respective set of colloquial-speaker vectors may be extracted. A matching procedure, carried out under a transform that compensates for speaker differences, may be used to match each colloquial-speaker vector to a reference-speaker vector. The colloquial-speaker vector may be replaced with the matched reference-speaker vector. The matching-and-replacing can be carried out separately for each set of colloquial-speaker vectors. A conditioned set of speaker vectors can then be constructed by aggregating all the replaced speaker vectors. The condition set of speaker vectors can be used to train the TTS system.

Abstract: One embodiment of the present invention provides a system for providing context-based web services for a user. During operation, the system receives a sentence as input from a user. The system performs natural language processing on the sentence to determine one or more parameters. The system retrieves data from a foreground knowledge graph containing contextual data for the user and from a background knowledge graph containing background information corresponding to the parameters. The system determines a set of arguments based on the parameters and/or data from the foreground knowledge graph and/or data from the background knowledge graph. The system then selects an action module based on results of the natural language processing and/or the set of arguments. The system passes the arguments to the action module. The action module then uses the arguments to respond to a question or interact with web services to perform an action for the user.

Abstract: Some embodiments of an efficient string search have been presented. In one embodiment, a string of bytes representing content written in a non-delimited language is received, wherein the content has been classified into a predetermined category. In a single pass through the string of bytes, a set of N-grams is searched for simultaneously. Statistical information on occurrences of the N-grams, if any, in the string of bytes is collected. In some embodiments, a model is generated based on the statistical information, where the model is usable by a content filter to classify content.

Abstract: In speech recognition, the duration of a phoneme is taken into account when determining recognition scores. Specifically, the duration of a phoneme may be evaluated relative to the duration of neighboring phonemes. A phoneme that is interpreted to be significantly longer or shorter than its neighbors may be given a lower duration score. A duration score for a phoneme may be calculated and used to adjust a recognition score. In this manner a duration model may supplement an acoustic model and language model to improve speech recognition results.

Abstract: A method for operating an electronic device is provided. The method includes determining one or more images; determining at least one first sound sources; dividing the first sound source into a plurality of second sound sources; and inserting at least one of the plurality of the second sound sources into the one or more images.

Abstract: A system and method of creating a customized multi-media message to a recipient is disclosed. The multi-media message is created by a sender and contains an animated entity that delivers an audible message. The sender chooses the animated entity from a plurality of animated entities. The system receives a text message from the sender and receives a sender audio message associated with the text message. The sender audio message is associated with the chosen animated entity to create the multi-media message. The multi-media message is delivered by the animated entity using as the voice the sender audio message wherein the mouth movements of the animated entity conform to the sender audio message.

Abstract: A system for providing an acoustic environment recognizer for optimal speech processing is disclosed. In particular, the system may utilize metadata obtained from various acoustic environments to assist in suppressing ambient noise interfering with a desired audio signal. In order to do so, the system may receive an audio stream including an audio signal associated with a user and including ambient noise obtained from an acoustic environment of the user. The system may obtain first metadata associated with the ambient noise, and may determine if the first metadata corresponds to second metadata in a profile for the acoustic environment. If the first metadata corresponds to the second metadata, the system may select a processing scheme for suppressing the ambient noise from the audio stream, and process the audio stream using the processing scheme. Once the audio stream is processed, the system may provide the audio stream to a destination.

Abstract: An apparatus and method for concealing frame erasure and a voice decoding apparatus and method using the same. The frame erasure concealment apparatus includes: a parameter extraction unit determining whether there is an erased frame in a voice packet, and extracting an excitement signal parameter and a line spectrum pair parameter of a previous good frame; and an erasure frame concealment unit, if there is an erased frame, restoring the excitement signal and line spectrum pair parameter of the erased frame by using a regression analysis from the excitement signal and line spectrum pair parameter of the previous good frame. According to the method and apparatus, by predicting and restoring the parameter of the erased frame through the regression analysis, the quality of the restored voice signal can be enhanced and the algorithm can be simplified.