Abstract: An object of embodiments of the present invention is to provide an improved gain-shape VQ. This is achieved by determining a number of bits to be allocated to a gain adjustment- and shape-quantizer for a plurality of combinations of a current bit rate and a first signal property. The determined allocated number of bits to the gain adjustment- and shape quantizer should provide a better result for the given bitrate and signal property than using a single fixed allocation scheme. That can be achieved by deriving the bit allocation by using an average of optimal bit allocations for a training data set. Thus by pre-calculating a number of bits to the gain adjustment and the shape quantizers for a plurality of combinations of the bit rate and a first signal property and creating a table indicating the number of bits to be allocated to the gain adjustment- and the shape-quantizers for a plurality of combinations of the bit rate and a first signal property.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for designating certain voice commands as hotwords. The methods, systems, and apparatus include actions of receiving a hotword followed by a voice command. Additional actions include determining that the voice command satisfies one or more predetermined criteria associated with designating the voice command as a hotword, where a voice command that is designated as a hotword is treated as a voice input regardless of whether the voice command is preceded by another hotword. Further actions include, in response to determining that the voice command satisfies one or more predetermined criteria associated with designating the voice command as a hotword, designating the voice command as a hotword.

Abstract: A speaker indexing device extracts a plurality of features from a speech signal on a frame-by-frame basis, models a distribution of first feature sets by a mixture distribution containing as many probability distributions as there are speakers, selects for each probability distribution either first feature sets located within a predetermined distance from the center of the probability distribution or a predetermined number of first feature sets in sequence starting from the first feature set closest to the center of the probability distribution, selects a second feature for the frame corresponding to the selected first feature sets as first training data for the speaker corresponding to the probability distribution and, using the first training data, trains a speaker model to be used to append to each frame identification information for identifying the speaker speaking in the frame.

Abstract: According to a preferred aspect of the instant invention, there is provided a system and method for content recognition in portable devices. Content, preferably audio content is recorded by the instant invention, preferably a sample with a length between 1 and 10 seconds. A fingerprint will be generated from the recorded sample and automatically, and preferably without further user interaction, prompting, notification, etc. (e.g. invisible to the user), compared with the fingerprints in a fingerprint database that is stored locally in the portable device and the result thereafter presented to the user.

Abstract: A method, an apparatus, and a computer-readable medium configured with instructions that when executed carry out the method for determining a measure of harmonicity. In one embodiment the method includes selecting candidate fundamental frequencies within a range, and for candidate determining a mask or retrieving a pre-calculated mask that has positive value for each frequency that contributed to harmonicity, and negative value for each frequency that contributes to inharmonicity. A candidate harmonicity measure is calculated for each candidate fundamental by summing the product of the mask and the magnitude measure spectrum. The harmonicity measure is selected as the maximum of the candidate harmonicity measures.

Abstract: The present invention is a method and apparatus for narrative content generation using narrative frameworks by receiving a first phrase variation and a second phrase variation and displaying an error indication when the first phrase variation fails to satisfy a criterion relative to the second phrase variation. If there is an error indication, alternate phrase variations are received and compared against the first phrase variation until an alternate phrase variation is selected that has no error indication. Additionally, multiple sets of operators for updating one or more narrative phrases selected for inclusion in the narrative content framework may be utilized to update selected phrases after inclusion in the narrative framework but prior to finalizing the narrative content to be output.

Abstract: A method for generating training data for disambiguation of an entity comprising a word or word string related to a topic to be analyzed includes acquiring sent messages by a user, each including at least one entity in a set of entities; organizing the messages and acquiring sets, each containing messages sent by each user; identifying a set of messages including different entities, greater than or equal to a first threshold value, and identifying a user corresponding to the identified set as a hot user; receiving an instruction indicating an object entity to be disambiguated; determining a likelihood of co-occurrence of each keyword and the object entity in sets of messages sent by hot users; and determining training data for the object entity on the basis of the likelihood of co-occurrence of each keyword and the object entity in the sets of messages sent by the hot users.

Abstract: A computer-implemented method and an apparatus for modeling customer interaction experiences receives interaction data corresponding to one or more interactions between a customer and a customer support representative. At least one language associated with the interaction data is detected. Textual content in a plurality of languages is generated corresponding to the interaction data based at least in part on translating the interaction data using two or more languages different than the at least one language. At least one emotion score is determined for text corresponding to each language from among the plurality of languages. An aggregate emotion score is determined using the at least one emotion score for the text corresponding to the each language. An interaction experience of the customer is modeled based at least in part on the aggregate emotion score.

Abstract: A first speech processing device includes a first speech input unit and a first speech output unit. A second speech processing device includes a second speech input unit and a second speech output unit. In a server therebetween, a speech of a first language sent from the first speech input unit is recognized. The speech recognition result is translated into a second language. The translation result is back translated into the first language. A first speech synthesis signal of the back translation result is sent to the first speech output unit. A second speech synthesis signal of the translation result is sent to the second speech output unit. Duration of the second speech synthesis signal or the first speech synthesis signal is measured. The first speech synthesis signal and the second speech synthesis signal are outputted by synchronizing a start time and an end time thereof, based on the duration.

Abstract: A system for verifying factual assertions in natural language. Receiving a text input. Identifying, a verifiable factual statement in the text. Forming a query based on the verifiable factual assertion. Searching a corpus based on the query. Determining the veracity of verifiable factual assertion based on the search results.

Abstract: An interactive response system directs input to a software-based router, which is able to intelligently respond to the input by drawing on a combination of human agents, advanced recognition and expert systems. The system utilizes human “intent analysts” for purposes of interpreting customer input. Automated recognition subsystems are trained by coupling customer input with IA-selected intent corresponding to the input, using model-updating subsystems to develop the training information for the automated recognition subsystems.

Abstract: According to one embodiment, a speech of a first language is recognized using a speech recognition dictionary to recognize the first language and a second language, and a source sentence of the first language is generated. The source sentence is translated into a second language, and a translation sentence of the second language is generated. An unknown word included in the translation sentence is detected. The unknown word is not stored in the speech recognition dictionary. A first pronunciation candidate of the unknown word is estimated, from a representation of the unknown word. A second pronunciation candidate of the unknown word is estimated from a pronunciation of an original word included in the source sentence corresponding to the unknown word. The unknown word, the first pronunciation candidate and the second pronunciation candidate, are registered into the speech recognition dictionary correspondingly.

Abstract: A method and associated system and computer program product. A sample of speech, for which a source of the sample of speech is to be classified, is received. A frequency clip level of the sample of speech is determined. A higher frequency clip level indicates the source is human and a lower frequency clip level indicates the source is machine generated. A dynamic range of the sample of speech is determined. A lower dynamic range indicates the source is human and a higher dynamic range indicates the source is machine generated. The frequency clip level and the dynamic range are weighted by a respective weighting factor as to whether the source is human or the source is machine generated. The source is classified as human generated or machine generated. The classifying of the source is based on the frequency clip level, the dynamic range, and the respective weighting factors thereof.

Abstract: In a speech processing apparatus, an acquisition unit is configured to acquire a speech. A separation unit is configured to separate the speech into a plurality of sections in accordance with a prescribed rule. A calculation unit is configured to calculate a degree of similarity in each combination of the sections. An estimation unit is configured to estimate, with respect to the each section, a direction of arrival of the speech. A correction unit is configured to group the sections whose directions of arrival are mutually similar into a same group and correct the degree of similarity with respect to the combination of the sections in the same group. A clustering unit is configured to cluster the sections by using the corrected degree of similarity.

Abstract: Disclosed herein are systems, computer-implemented methods, and computer-readable storage media for generating personalized user models. The method includes receiving automatic speech recognition (ASR) output of speech interactions with a user, receiving an ASR transcription error model characterizing how ASR transcription errors are made, generating guesses of a true transcription and a user model via an expectation maximization (EM) algorithm based on the error model and the respective ASR output where the guesses will converge to a personalized user model which maximizes the likelihood of the ASR output. The ASR output can be unlabeled. The method can include casting speech interactions as a dynamic Bayesian network with four variables: (s), (u), (r), (m), and encoding relationships between (s), (u), (r), (m) as conditional probability tables. At each dialog turn (r) and (m) are known and (s) and (u) are hidden.

Abstract: A pre-processing apparatus for speech recognition may include: a trailing silence period detection unit configured to detect the length of a trailing silence period contained in a speech signal; a reference trailing silence period storage unit configured to store the length of a reference trailing silence period; and a trailing silence period adjusting unit configured to adjust the length of the trailing silence period contained in the speech signal based on the length of the reference trailing silence period.

Abstract: A method of extracting the semantic distance from the mathematical sentence and classifying the mathematical sentence by the semantic distance, includes: receiving a user query; extracting at least one keyword included in the received user query; and extracting a semantic distance by, indexing one or more of natural language tokens and mathematical equation tokens including semantic information, extracting the semantic distance, between the at least one extracted keyword and the one or more indexed semantic information by referring indexed information, and acquiring a similarity of the received user query and the semantic information.

Abstract: An electronic device includes a microphone that receives an audio signal, and a processor that is electrically coupled to the microphone. The processor detects a trigger phrase in the received audio signal and measure characteristics of the detected trigger phrase. Based on the measured characteristics of the detected trigger phrase, the processor determines whether the detected trigger phrase is valid.

Abstract: A method, apparatus, and system relating to embedding hidden content within a Unicode message and using the hidden content to perform a particular computer action.

Abstract: An automobile to detect a signal, having a security code, from a mobile transponder within range of the automobile is disclosed. The automobile may determine that the security code is valid to process an audible command from a user. The audible command may correspond to an automobile function. The automobile may also determine if the audible command matches a voiceprint of the user and process the automobile function accordingly. The automobile function may provide customized user settings for utilization of the automobile.