Abstract: A method for adaptive voice interaction includes monitoring voice communications between a service recipient and a service representative, measuring a set of voice communication features based upon the voice communications between the service recipient and the service representative, analyzing the set of voice communication features to generate emotion metric values, and generating a response based on the analysis of the set of voice communication features.

Abstract: A machine translation apparatus configured to translate an input sentence and output a translated sentence in a target language, the machine translation apparatus includes a rule acquirer configured to acquire a difference between an input example sentence and a replaced example sentence which is obtained by replacing the input example sentence, and acquire a replacement rule based on the difference and each of meaning representations which indicate each relationship of words in the input example sentence and the replaced example sentence; and a translator configured to apply the replacement rule acquired by the rule acquirer to the input sentence and output the translated sentence based on the meaning representations and a meaning representation indicating of relationship of words in the input sentence.

Abstract: A system and method are disclosed that improve automatic speech recognition in a spoken dialog system. The method comprises partitioning speech recognizer output into self-contained clauses, identifying a dialog act in each of the self-contained clauses, qualifying dialog acts by identifying a current domain object and/or a current domain action, and determining whether further qualification is possible for the current domain object and/or current domain action. If further qualification is possible, then the method comprises identifying another domain action and/or another domain object associated with the current domain object and/or current domain action, reassigning the another domain action and/or another domain object as the current domain action and/or current domain object and then recursively qualifying the new current domain action and/or current object. This process continues until nothing is left to qualify.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for speaker verification. The methods, systems, and apparatus include actions of inputting speech data that corresponds to a particular utterance to a first neural network and determining an evaluation vector based on output at a hidden layer of the first neural network. Additional actions include obtaining a reference vector that corresponds to a past utterance of a particular speaker. Further actions include inputting the evaluation vector and the reference vector to a second neural network that is trained on a set of labeled pairs of feature vectors to identify whether speakers associated with the labeled pairs of feature vectors are the same speaker. More actions include determining, based on an output of the second neural network, whether the particular utterance was likely spoken by the particular speaker.

Abstract: Audio watermarking is the process of embedding watermark information items into an audio signal in an in-audible manner. In a first embodiment, in case the original audio signal has parts of low signal energy, an alternative signal having a level or strength given by the psycho-acoustic model is combined with the original audio signal. The combined signal is watermarked with watermark data to be embedded. In a second embodiment, in case the original audio signal has parts of low signal energy, an alternative signal having a level or strength given by the psycho-acoustic model is watermarked with watermark data to be embedded, and the audio signal is watermarked with the watermark data to be embedded. The watermarked alternative signal is combined with the watermarked audio signal.

Abstract: On a computing device a speech utterance is received from a user. The speech utterance is a section of a speech dialog that includes a plurality of speech utterances. One or more features from the speech utterance are identified. Each identified feature from the speech utterance is a specific characteristic of the speech utterance. One or more features from the speech dialog are identified. Each identified feature from the speech dialog is associated with one or more events in the speech dialog. The one or more events occur prior to the speech utterance. One or more identified features from the speech utterance and one or more identified features from the speech dialog are used to calculate a confidence score for the speech utterance.

Abstract: Systems and methods of performing blind bandwidth extension are disclosed. In an embodiment, a method includes determining, based on a set of low-band parameters of an audio signal, a first set of high-band parameters and a second set of high-band parameters. The method further includes generating a predicted set of high-band parameters based on a weighted combination of the first set of high-band parameters and the second set of high-band parameters.

Abstract: A voice recognition system that divides a search space for voice recognition into a general domain search space and a specific domain search space. A mobile terminal receives a voice recognition target word from a user, and a voice recognition server divides a search space for voice recognition into a general domain search space and a specific domain search space and stores the spaces and performs voice recognition for the voice recognition target word through linkage of the general domain search space and the specific domain search space.

Abstract: A custom dictionary is generated for an e-book. A dictionary management system receives a custom dictionary request from a user client operated by a user, the custom dictionary request identifying the e-book and including dictionary management information describing the user. The dictionary management system chooses a group reader profile that has an associated group reading score for the user based on the dictionary management information and candidate words are identified in the identified e-book for inclusion in the custom dictionary. The dictionary management system selects words for inclusion in the custom dictionary from among the candidate words responsive to the associated group reading score for the chosen group reading profile. The dictionary management system generates the custom dictionary using the selected words, and provides the generated custom dictionary to the user client.

Abstract: The present invention relates to an audio signal coding method and apparatus. The method includes: categorizing audio signals into high-frequency audio signals and low-frequency audio signals; coding the low-frequency audio signals by using a corresponding low-frequency coding manner according to characteristics of low-frequency audio signals; and selecting a bandwidth extension mode to code the high-frequency audio signals according to the low-frequency coding manner and/or characteristics of the audio signals.

Abstract: A method for updating language understanding classifier models includes receiving via one or more microphones of a computing device, a digital voice input from a user of the computing device. Natural language processing using the digital voice input is used to determine a user voice request. Upon determining the user voice request does not match at least one of a plurality of pre-defined voice commands in a schema definition of a digital personal assistant, a GUI of an end-user labeling tool is used to receive a user selection of at least one of the following: at least one intent of a plurality of available intents and/or at least one slot for the at least one intent. A labeled data set is generated by pairing the user voice request and the user selection, and is used to update a language understanding classifier.

Abstract: An audio encoder has a window function controller, a windower, a time warper with a final quality check functionality, a time/frequency converter, a TNS stage or a quantizer encoder, the window function controller, the time warper, the TNS stage or an additional noise filling analyzer are controlled by signal analysis results obtained by a time warp analyzer or a signal classifier. Furthermore, a decoder applies a noise filling operation using a manipulated noise filling estimate depending on a harmonic or speech characteristic of the audio signal.

Abstract: In general, techniques are described for coding an ambient higher order ambisonic coefficient. An audio decoding device comprising a memory and a processor may perform the techniques. The memory may store a first frame of a bitstream and a second frame of the bitstream. The processor may obtain, from the first frame, one or more bits indicative of whether the first frame is an independent frame that includes additional reference information to enable the first frame to be decoded without reference to the second frame. The processor may further obtain, in response to the one or more bits indicating that the first frame is not an independent frame, prediction information for first channel side information data of a transport channel. The prediction information may be used to decode the first channel side information data of the transport channel with reference to second channel side information data of the transport channel.

Abstract: In an embodiment, bitstream elements of sub-frames are encoded differentially to a global gain value so that a change of the global gain value results in an adjustment of an output level of the decoded representation of the audio content. Concurrently, the differential coding saves bits. Even further, the differential coding enables the lowering of the burden of globally adjusting the gain of an encoded bitstream. In another embodiment, a global gain control across CELP coded frames and transform coded frames is achieved by co-controlling the gain of the codebook excitation of the CELP codec, along with a level of the transform or inverse transform of the transform coded frames. In another embodiment, the gain value determination in CELP coding is performed in the weighted domain of the excitation signal.

Abstract: Disclosed is a method and apparatus for responding to an inquiry from a client via a network. The method and apparatus receive the inquiry from a client via a network. Based on the inquiry, question-answer pairs retrieved from the network are analyzed to determine a response to the inquiry. The QA pairs are not predefined. As a result, the QA pairs have to be analyzed in order to determine whether they are responsive to a particular inquiry. Questions of the QA pairs may be repetitive and, without more, will not be useful in determining whether their corresponding answer responds to an inquiry.

Abstract: In general, techniques are described for indicating frame parameter reusability for decoding vectors. A device comprising a processor and a memory may perform the techniques. The processor may be configured to obtain a bitstream comprising a vector representative of an orthogonal spatial axis in a spherical harmonics domain. The bitstream may further comprise an indicator for whether to reuse, from a previous frame, at least one syntax element indicative of information used when compressing the vector. The memory may be configured to store the bitstream.

Abstract: A computing system is configured to listen to user speech and translate the user speech into voice commands that control operation of the computing system. The identity of a user interacting with the computing system is determined, and a voice command is selected from a set of voice commands based on the user identity. A voice-command suggestion corresponding to the voice command is selected and presented via a display. If the user speaks the voice-command suggestion, the computing system executes the voice command corresponding to the voice-command suggestion.

Abstract: According to an embodiment, a speech synthesis dictionary generation apparatus includes an analyzer, a speaker adapter, a level designation unit, and a determination unit. The analyzer analyzes speech data and generates a speech database containing characteristics of utterance by an object speaker. The speaker adapter generates the model of the object speaker by speaker adaptation of converting a base model to be closer to characteristics of the object speaker based on the database. The level designation unit accepts designation of a target speaker level representing a speaker's utterance skill and/or a speaker's native level in a language of the speech synthesis dictionary. The determination determines a parameter related to fidelity of reproduction of speaker properties in the speaker adaptation, in accordance with a relationship between the target speaker level and a speaker level of the object speaker.

Abstract: Disclosed are techniques and systems to provide a narration of a text. In some aspects, the techniques and systems described herein include generating a timing file that includes elapsed time information for expected portions of text that provides an elapsed time period from a reference time in an audio recording to each portion of text in recognized portions of text.

Abstract: An automated system to build a user audio profile for a natural or continuous language speech to text dictation/transcription system is provided. The system uses previously recorded audio files that may have been already transcribed. The previously recorded audio file is split into a plurality of smaller audio files of about 15 seconds in length. The plurality of smaller audio files are matched to the transcribed text (e.g., small text files) or the smaller audio files are transcribed. All, some, or a selection of the small audio files and the small text files are linked as a training pair. The training pair may be edited in certain embodiments herein, both the text and the audio. The training pairs are submitted to the server to build the initial user audio profile.