Abstract: According to an embodiment of a present disclosure, a method for generating and utilizing a text fingerprint performed by a computing device is disclosed. The method comprises the steps of: dividing text data into one or more segments based on a predetermined text segmentation algorithm; determining a mapping value assigned to one or more subsegments that form the divided segment based on a predetermined mapping algorithm; generating a coordinate value for each of the one or more segments based on the determined mapping value; and generating the virtual fingerprint having a phonetic feature for the text data based on the generated coordinate value. That is, whether a plurality of the text data are similar to each other can be easily determined, when the text data also has a unique virtual fingerprint based on a pronunciation, in the same way every person has a unique fingerprint.

Abstract: An electronic apparatus and a control method thereof are provided. A method of controlling an electronic apparatus according to an embodiment of the disclosure includes: receiving input of a first utterance, identifying a first task for the first utterance based on the first utterance, providing a response to the first task based on a predetermined response pattern, receiving input of a second utterance, identifying a second task for the second utterance based on the second utterance, determining the degree of association between the first task and the second task, and setting a response pattern for the first task based on the second task based on the determined degree of association satisfying a predetermined condition. The control method of an electronic apparatus may use an artificial intelligence model trained according to at least one of machine learning, a neural network, or a deep learning algorithm.

Abstract: A method for generating a set of concepts related to a target concept includes accessing a set of candidate concepts, embedding the target concept and the set of candidate concepts in a semantic vector space, selecting one or more intermediate concepts from the set of candidate concepts in response to determining whether each embedded candidate concept in the set of embedded candidate concepts satisfies a predetermined relationship with the embedded target concept, and filtering the one or more intermediate concepts to yield the set of concepts related to the target concept. The method may further include generating a multiple-choice question in which the target concept corresponds to a correct answer choice and the set of concepts related to the target concept correspond to distractors.

Abstract: A voice identity feature extractor training method includes extracting a voice feature vector of training voice, The method may include determining a corresponding I-vector according to the voice feature vector of the training voice. The method may include adjusting a weight of a neural network model by using the I-vector as a first target output of the neural network model, to obtain a first neural network model, The method may include obtaining a voice feature vector of target detecting voice and determining an output result of the first neural network model for the voice feature vector of the target detecting voice. The method may include determining an I-vector latent variable. The method may include estimating a posterior mean of the I-vector latent variable, and adjusting a weight of the first neural network model using the posterior mean as a second target output, to obtain a voice identity feature extractor.

Abstract: Provided are techniques for domain specific correction of output from automatic speech recognition. An output of an automatic speech recognition engine is received. An alphanumeric sequence is extracted from the output, where the alphanumeric sequence represents an erroneous translation by the automatic speech recognition engine. Candidates for the alphanumeric sequence are generated. The candidates are ranked based on scores associated with the candidates. A candidate of the candidates having a highest score of the scores is selected. The output is corrected by replacing the alphanumeric sequence with the selected candidate. The corrected output is returned.

Abstract: Embodiments generally relate to the generation of a domain-specific phrasal dictionary. In some embodiments, a method includes receiving text from a user, wherein the text includes unstructured text of a natural language. The method further includes parsing the text into text chunks. The method further includes sending the text chunks to the user. The method further includes receiving one or more phrase categories and one or more predetermined phrases from the user, wherein each predetermined phrase of the one or more predetermined phrases corresponds to at least one phrase category of the one or more phrase categories. The method further includes comparing the predetermined phrases with the text chunks. The method further includes assigning at least one phrase category of the one or more phrase categories to at least one text chunk. The method further includes sending at least one text chunk and the at least one phrase category that is assigned to the at least one text chunk to the user.

Abstract: Provided is a system for generating speech recognition training data, the system including: a speech data processing module receiving speech data from a user terminal and performing data preprocessing on the received speech data; an auto speech recognition (ASR) interfacing module transmitting the preprocessed speech data to a plurality of ASR engines and acquiring a confidence score and transcription data of the speech data from the plurality of ASR engines; an ASR result evaluating module determining whether the speech data and the transcription data match each other; and a training data managing unit generating training data as a pair of the speech data and the transcription data determined to match each other.

Abstract: A method includes receiving text input data for conversion into synthesized speech and determining, using a hotword-aware model trained to detect a presence of a hotword assigned to a user device, whether a pronunciation of the text input data includes the hotword. The hotword is configured to initiate a wake-up process on the user device for processing the hotword and/or one or more other terms following the hotword in the audio input data. When the pronunciation of the text input data includes the hotword, the method also includes generating an audio output signal from the text input data and providing the audio output signal to an audio output device to output the audio output signal. The audio output signal when captured by an audio capture device of the user device, configured to prevent initiation of the wake-up process on the user device.

Abstract: A system and method performs wakeword detection and automatic speech recognition using the same acoustic model. A mapping engine maps phones/senones output by the acoustic model to phones/senones corresponding to the wakeword. A hidden Markov model (HMM) may determine that the wakeword is present in audio data; the HMM may have multiple paths for multiple wakewords or may have multiple models. Once the wakeword is detected, ASR is performed using the acoustic model.

Abstract: Apparatus and methods for interactive voice recognition. The apparatus and methods may include a canonical phrase derivation engine configured to derive canonical phrases from voice data. The apparatus may include an input engine configured to parse utterances. The apparatus may include a knowledge extraction engine to disambiguate the utterances into words, form a sequence from the words, extract context from the sequence, pair the sequence with a phrase of the canonical phrases, merge the sequence and the phrase to form a hybrid phrase, vectorize the hybrid phrase into a vector, and feed the vector into a non-linear classification engine to determine an intent corresponding to the utterances.

Abstract: The present disclosure provides various embodiments of methods for intelligent active speaker identification and information handling systems (IHSs) utilizing such methods. In general, the methods disclosed herein may be used to accurately identify an active speaker in a communication session with an application or an IHS, regardless of whether the active speaker is alone, in a group environment, or using someone else's system or login to participate in the communication session. The methods disclosed herein may use voice processing technology and one or more voice identification databases (VIDs) to identify the active speaker in a communication session. In some embodiments, the disclosed methods may display the identity of the active speaker to other users or participants in the same communication session. In other embodiments, the disclosed methods may dynamically switch between user profiles or accounts during the communication session based on the identity of the active speaker.

Abstract: A method with registration includes: receiving a speech signal of a speaker; synthesizing the received speech signal and a noise signal to generate a synthesized signal; generating a feature vector based on the synthesized signal; and constructing a registration database (DB) corresponding to the speaker based on the generated feature vector.

Abstract: Implemented are an apparatus and a method that enable highly accurate intent estimation of a user utterance. An utterance learning adaptive processing unit analyzes a plurality of user utterances input from a user, generates learning data in which entity information included in a user utterance with an unclear intent is associated with a correct intent, and stores the generated learning data is a storage unit. The utterance learning adaptive processing unit generates learning data in which an intent, acquired from a response utterance from the user to an apparatus utterance after input of a first user utterance with an unclear intent, is recorded in association with entity information included in the first user utterance. The learning data is recorded to include superordinate semantic concept information of the entity information. At the time of estimating an intent for a new user utterance, learning data with similar superordinate semantic concept information is used.

Abstract: A method includes identifying a set of identigens for each word of a first phrase of a phrase group to produce a first plurality of sets of identigens and determining whether first and second identigen rules are applicable to the first plurality of sets of identigens. When the first and second identigen rules are applicable to the first plurality of sets of identigens, the method further includes identifying a set of identigens for each word of a second phrase of the phrase group to produce a second plurality of sets of identigens and determining that the first identigen rules are applicable to the second plurality of sets of identigens. The method further includes identifying one valid identigen of each set of identigens of the first plurality of sets of identigens by applying the first identigen rules to the first plurality of sets of identigens to produce a first entigen group.

Abstract: A plurality of abbreviated names are generated for evaluation based on a full name of a point-of-interest (POI) on a map. A plurality of address names comprising the full name of the POI or any of the abbreviated names to be evaluated are obtained from a predetermined area of the POI. A phrase status vector used to indicate a location status of the target phrase in each particular address name is calculated for a target phrase based on each address name, the target phrase including the full name of the POI or any of the abbreviated names to be evaluated. A similarity is calculated between a phrase status vector for the full name of the POI and a phrase status vector. A particular abbreviated name corresponding with a calculated similarity greater than a predetermined threshold is associated with the full name of the POI corresponding to the calculated similarity.

Abstract: A system and method for detecting providing a multi-region software as a service model in a cloud computing system, including for example, obtaining a list of allowed regions that can host data of a user from a source region; and automatically assigning a device associated with a new user from the source region to a computing system located in one of the allowed regions. The source region may be determined based on location information obtained from the device during registration.

Abstract: A method includes monitoring a plurality of system inputs, and detecting a behavioral pattern performed by a user and associated with the plurality of system inputs, When the behavioral pattern is detected, the method includes associating, in a memory, a gesture with at least one action, the at least one action being determined by the plurality of system inputs, and, upon detecting the gesture, executing the action associated with the gesture.

Abstract: Proposed are a generative adversarial network-based speech bandwidth extender and extension method. A generative adversarial network-based speech bandwidth extension method, according to an embodiment, comprises the steps of: extracting feature vectors from a narrowband (NB) signal and a wideband (WB) signal of a speech; estimating the feature vector of the wideband signal from the feature vector of the narrowband signal; and learning a deep neural network classification model for discriminating the estimated feature vector of the wideband signal from the actually extracted feature vector of the wideband signal and the actually extracted feature vector of the narrowband signal.

Abstract: Systems and methods to improve the performance of an automatic speech recognition (ASR) system using a confusion index indicative of the amount of confusion between words are described, where a confusion index (CI) or score is calculated by receiving a first word (Word1) and a second word (Word2), calculating an acoustic score (A12) indicative of the phonetic difference between Word1 and Word2, calculating a weighted language score (W(U1+U2), indicative of a weighted likelihood (or word frequency) of Word1 and Word2 occurring in the corpus, the confusion index CI incorporating both the acoustic score and the weighted language score, such that the CI for words that sound alike and have a high likelihood of occurring in the corpus will be higher than the CI for words that sound alike and do not have a high likelihood of occurring in the corpus.

Abstract: A method for predicting parametric vocoder parameter includes receiving a text utterance having one or more words, each word having one or more syllables, and each syllable having one or more phonemes. The method also includes receiving, as input to a vocoder model, prosodic features that represent an intended prosody for the text utterance and a linguistic specification. The prosodic features include a duration, pitch contour, and energy contour for the text utterance, while the linguistic specification includes sentence-level linguistic features, word-level linguistic features for each word, syllable-level linguistic features for each syllable, and phoneme-level linguistic features for each phoneme. The method also includes predicting vocoder parameters based on the prosodic features and the linguistic specification.

Abstract: The present invention provides a device for estimating the deterioration factor of speech recognition accuracy by estimating an acoustic factor that leads to a speech recognition error. The device extracts an acoustic feature amount for each frame from an input speech, calculates a posterior probability for each acoustic event for the acoustic feature amount for each frame, corrects the posterior probability by filtering the posterior probability for each acoustic event using a time-series filter with weighting coefficients developed in the time axis, outputs a set of speech recognition results with a recognition score, outputs a feature amount for the speech recognition results for each frame, calculates and outputs a principal deterioration factor class for the speech recognition accuracy for each frame on the basis of the corrected posterior probability, the feature amount for speech recognition results for each frame, and the acoustic feature amount for each frame.

Abstract: System and method for generating disambiguated terms in automatically generated transcriptions including instructions within a knowledge domain and employing the system are disclosed.

Abstract: Described herein is a system for prompting a user for clarification when an automatic speech recognition (ASR) system encounters ambiguity with respect to the user's input. The feedback provided by the user is used to retrain machine-learning models and/or to generate new machine-learning models. Based on the type of ambiguity, the system may determine to retrain one or more ASR models that are widely used by the system or to generate/update one or more user-specific models that are used to process inputs from one or more particular users.

Abstract: Provided is an artificial intelligence apparatus for predicting a performance of a voice recognition model in a user environment including: a memory configured to store a performance prediction model; and a processor configured to: obtain first controlled environment data including first controlled environment factors corresponding to a first controlled voice recognition environment and a first controlled voice recognition performance of a target voice recognition model in the first controlled voice recognition environment; obtain first user environment factors corresponding to a first user environment, in which the performance is to be predicted; predict, using the performance prediction model, a first user voice recognition performance of the target voice recognition model in the first user voice recognition environment from the obtained first controlled environment data and the first user environment factors; and output the predicted first user voice recognition performance.

Abstract: A speech emotion detection system may obtain to-be-detected speech data. The system may generate speech frames based on framing processing and the to-be-detected speech data. The system may extract speech features corresponding to the speech frames to form a speech feature matrix corresponding to the to-be-detected speech data. The system may input the speech feature matrix to an emotion state probability detection model. The system may generate, based on the speech feature matrix and the emotion state probability detection model, an emotion state probability matrix corresponding to the to-be-detected speech data. The system may input the emotion state probability matrix and the speech feature matrix to an emotion state transition model. The system may generate an emotion state sequence based on the emotional state probability matrix, the speech feature matrix, and the emotional state transition model. The system may determine an emotion state based on the emotion state sequence.

Abstract: Language models may be run with an input set of words in a given sentence. Each of the language models can predict a set of next candidate words to follow the input set or words. Based on the sets of next candidate words predicted by the language models and an actual next word, language guidance can be provided.

Abstract: A system for distinguishing between a human voice generated command and an electronic speaker generated command is provided. An exemplary system comprises a microphone array for receiving an audio signal collection, preprocessing circuitry configured for converting the audio signal collection into processed recorded audio signals, energy balance metric determination circuitry configured for calculating a final energy balance metric based on the processed recorded audio signals, and energy balance metric evaluation circuitry for outputting a command originator signal based at least in part on the final energy balance metric.

Abstract: Aspects of the present invention disclose a method, computer program product, and system for modifying a prohibited list. The method includes one or more processors determining a trend in usage of a command based on an analysis of aggregated usage information of the command from a plurality of listening devices, wherein the determined trend indicates an amount of usage of the command over a period of time. The method further includes determining whether the determined trend in usage of the command deviates from a usage threshold for the command. The method further includes in response to determining that the determined trend in usage of the command does deviate from the usage threshold for the command, modifying a prohibited list based on an amount of deviation from the usage threshold usage of the command.

Abstract: In an approach to soft-forgetting training, one or more computer processors train a first model utilizing one or more training batches wherein each training batch of the one or more training batches comprises one or more blocks of information. The one or more computer processors, responsive to a completion of the training of the first model, initiate a training of a second model utilizing the one or more training batches. The one or more computer processors jitter a random block size for each block of information for each of the one or more training batches for the second model. The one or more computer processors unroll the second model over one or more non-overlapping contiguous jittered blocks of information. The one or more computer processors, responsive to the unrolling of the second model, reduce overfitting for the second model by applying twin regularization.

Abstract: Generally discussed herein are devices, systems, and methods for wake word verification. A method can include receiving, at a server, a message from a device indicating that an utterance of a user-defined wake word was detected at the device, the message including (a) audio samples or features extracted from the audio samples and (b) data indicating the user-defined wake word, retrieving or generating, at the server, a custom decoding graph for the user-defined wake word, wherein the decoding graph and the static portion of the wake word verification model form a custom wake word verification model for the user-defined wake word, executing the wake word verification model to determine a likelihood that the wake word was uttered, and providing a message to the device indicating whether wake was uttered based on the determined likelihood.

Abstract: The invention discloses a multi-language mixed speech recognition method, which belongs to the technical field of speech recognition; the method comprises: step S1, configuring a multi-language mixed dictionary including a plurality of different languages; step S2, performing training according to the multi-language mixed dictionary and multi-language speech data including a plurality of different languages to form an acoustic recognition model; step S3, performing training according to multi-language text corpus including a plurality of different languages to form a language recognition model; step S4, forming the speech recognition system by using the multi-language mixed dictionary, the acoustic recognition model and the language recognition model; and subsequently, recognizing mixed speech by using the speech recognition system, and outputting a corresponding recognition result.

Abstract: A method may include dividing input audio into frames and calculating a characteristic value for each of the frames. The method may include establishing a voting matrix having a first dimension representing a quantity of segments of sample audio and a second dimension representing a quantity of frames of each segment. The method may include marking voting labels in the voting matrix corresponding to frames of the sample audio when the characteristic values of corresponding frames of the input audio and sample audio match. The method may include determining a frame to be a recognition result when a sum of the voting labels at a corresponding position is higher than a threshold.

Abstract: An exemplary automatic speech recognition (ASR) system may receive an audio input including a segment of speech. The segment of speech may be independently processed by general ASR and domain-specific ASR to generate multiple ASR results. A selection between the multiple ASR results may be performed based on respective confidence levels for the general ASR and domain-specific ASR. As incremental ASR is performed, a composite result may be generated based on general ASR and domain-specific ASR.

Abstract: A query-processing server that interprets natural language expressions supports the extension of a first semantic grammar (for a particular type of expression), which is declared extensible, by a second semantic grammar (for another type of expression). When an extension is requested, the query-processing server checks that the two semantic grammars have compatible semantic types. The developers need not have any knowledge of each other, or about their respective grammars. Performing an extension may be done by yet another party, such as the query-processing server, or another server, independently of all previous parties. The use of semantic grammar extensions provides a way to expand the coverage and functionality of natural language interpretation in a simple and flexible manner, so that new forms of expression may be supported, and seamlessly combined with pre-existing interpretations. Finally, in some implementations, this is done without loss of efficiency.

Abstract: A method for training a voice data set is provided. A first test set of data selected from a first voice data set, and a first voice model parameter obtained by performing first voice model training based on a first voice data set, are obtained. Data from a second voice data set is randomly selected to generate a second test set. Further, second voice model training is performed based on the second voice data set and the first voice model parameter when the second test set and the first test set satisfy a similarity condition.

Abstract: The disclosed computer-implemented method for performing natural language translation in AR may include accessing an audio input stream that includes words spoken by a speaking user in a first language. The method may next include performing active noise cancellation on the words in the audio input stream so that the spoken words are suppressed before reaching a listening user. Still further, the method may include processing the audio input stream to identify the words spoken by the speaking user, and translating the identified words spoken by the speaking user into a second, different language. The method may also include generating spoken words in the second, different language using the translated words, and replaying the generated spoken words in the second language to the listening user. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: According to one embodiment, an information processing apparatus includes one or more processors configured to detect a trigger from a voice signal, the trigger indicating start of voice recognition; and to perform voice recognition of a recognition sound section subsequent to a trigger sound section including the detected trigger, referring to a trigger and voice recognition dictionary corresponding to the trigger.

Abstract: Techniques for linguistic personalization of messages for targeted campaigns are described. In one or more implementations, dependencies between keywords and modifiers are extracted, from one or more segment-specific texts and a product-specific text, to build language models for the one or more segment specific texts and the product specific text. Modifiers with a desired sentiment are extracted from the product specific text and transformation points are identified in a message skeleton. Then one or more of the extracted modifiers are inserted to modify one or more identified keywords in the message skeleton to create a personalized message for a target segment of the targeted marketing campaign.

Abstract: Methods and systems for a transportation vehicle are provided. One method includes receiving a user input for a valid communication session by a processor executable, digital assistant at a device on a transportation vehicle; tagging by the digital assistant, the user input words with a grammatical connotation; generating an action context, a filter context and a response context by a neural network, based on the tagged user input; storing by the digital assistant, a key-value pair for a parameter of the filter context at a short term memory, based on an output from the neural network; updating by the digital assistant, the key-value pair at the short term memory after receiving a reply to a follow-up request and another output from the trained neural network; and providing a response to the reply by the digital assistant.

Abstract: A speech recognition method includes obtaining captured voice information, and determining semantic information of the captured voice information; segmenting the captured voice information to obtain voice segments when the semantic information does not satisfy a preset rule, and extracting voiceprint information of the voice segments; obtaining an unmatched voiceprint information from a local voiceprint database; matching the voiceprint information of the voice segments with the unmatched voiceprint information to determine a set of filtered voice segments having voiceprint information that successfully matches the unmatched voiceprint information; combining the set of filtered voice segments to obtain combined voice information, and determining combined semantic information of the combined voice information; and using the combined semantic information as a speech recognition result when the combined semantic information satisfies the preset rule.

Abstract: A medicine management method includes acquiring user information of medicine usage corresponding to at least one assigned medicine; acquiring user information of medicine using reactions; and displaying the user information of medicine usage with first time information corresponding to different time points of using the at least one assigned medicine and the user information of medicine using reactions with second time information corresponding to different time points of occurring medicine using reactions concurrently.

Abstract: A pitch detection method. Such a pitch detection method may have M-PWVT-TEO algorithm to detect a pitch value from a speech signal, apply a partial auto-correlation to a current signal with the pitch value to compensate the delay of the pitch value. Also, the pitch detection method may apply a full auto-correlation to the speech signal where the pitch value is not detected to recover on-sets of the speech signal.

Abstract: Methods and systems are disclosed in which audio broadcasts are converted into audio segments, for example, based on segment content. These audio segments are indexed, so as to be searchable, as computer searchable segments, for example, by network search engines and other computerized search tools.

Abstract: Techniques are provided for reverberation compensation for far-field speaker recognition. A methodology implementing the techniques according to an embodiment includes receiving an authentication audio signal associated with speech of a user and extracting features from the authentication audio signal. The method also includes scoring results of application of one or more speaker models to the extracted features. Each of the speaker models is trained based on a training audio signal processed by a reverberation simulator to simulate selected far-field environmental effects to be associated with that speaker model. The method further includes selecting one of the speaker models, based on the score, and mapping the selected speaker model to a known speaker identification or label that is associated with the user.

Abstract: Systems, methods, and apparatus are disclosed for generating and processing natural language requests. A request processing system processes a received natural language request to identify an intent of the natural language request and a confidence level of the identified intent. In response to the confidence level of the identified intent not satisfying a threshold level, the request processing system sends the natural language request to the fulfillment computing device for further processing by a person associated with the fulfillment computing device. In response to the confidence level satisfying the threshold level, the request processing system proceeds with fulfilling the request per the identified intent.

Abstract: A system determines user intent from a received conversation element. A plurality of distinct intent labels are generated for the received conversation element. The generated plurality of distinct intent labels are divided into a plurality of interpretation partitions with overlapping semantic content. for each interpretation partition of the plurality of interpretation partitions, a set of maximal coherent subgroups are defined that do not disagree on labels for terms in each subgroup, a score is computed for each maximal coherent subgroup of the defined set of maximal coherent subgroups, and a maximal coherent subgroup is selected from the set of maximal coherent subgroups based on the computed score. Intent labels are aggregated from the selected maximal coherent subgroup of each interpretation partition of the plurality of interpretation partitions to define a multiple intent interpretation of the received conversation element.

Abstract: A method of updating speech recognition data including a language model used for speech recognition, the method including obtaining language data including at least one word; detecting a word that does not exist in the language model from among the at least one word; obtaining at least one phoneme sequence regarding the detected word; obtaining components constituting the at least one phoneme sequence by dividing the at least one phoneme sequence into predetermined unit components; determining information regarding probabilities that the respective components constituting each of the at least one phoneme sequence appear during speech recognition; and updating the language model based on the determined probability information.

Abstract: A voice command application allows a user to configure an infotainment system to respond to customized voice commands. The voice command application exposes a library of functions to the user which the infotainment system can execute via interaction with the vehicle. The voice command application receives a selection of one or more functions and then receives a speech sample of the voice command. The voice command application generates sample metadata that includes linguistic elements of the voice command, and then generates a command specification. The command specification indicates the selected functions and the sample metadata for storage in a database. Subsequently, the voice command application receives the voice command from the user and locates the associated command specification in the database. The voice command application then extracts the associated set of functions and causes the vehicle to execute those functions to perform vehicle operations.

Abstract: A method for controlling operation of a power switch includes obtaining, by one or more processors of a power switch, data indicative of one or more non-contact gestures. The method includes determining, by the one or more processors, a control action based at least in part on the data indicative of the one or more non-contact gestures. The method includes implementing, by the one or more processors, the control action.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for computing numeric representations of words. One of the methods includes obtaining a set of training data, wherein the set of training data comprises sequences of words; training a classifier and an embedding function on the set of training data, wherein training the embedding function comprises obtained trained values of the embedding function parameters; processing each word in the vocabulary using the embedding function in accordance with the trained values of the embedding function parameters to generate a respective numerical representation of each word in the vocabulary in the high-dimensional space; and associating each word in the vocabulary with the respective numeric representation of the word in the high-dimensional space.