Abstract: A system and method for summarizing video feeds correlated to identified speakers. A transcriber system includes multiple types of reasoning logic for identifying specific speakers contained in the video feed. Each type of reasoning logic is stored in memory and may be combined and configurable to provide an aggregated speaker identification result useful for full or summarized transcription before transmission across a network for display on a network-accessible device.

Abstract: Techniques for naming devices via voice commands are described herein. For instance, a user may issue a voice command to a voice-controlled device stating, “you are the kitchen device”. Thereafter, the device may respond to voice commands directed, by name, to this device. For instance, the user may issue a voice command requesting to “play music on my kitchen device”. Given that the user has configured the device to respond to this name, the device may respond to the command by outputting the requested music.

Abstract: A computing system receives requests from client devices to process voice queries that have been detected in local environments of the client devices. The system identifies that a value that is based on a number of requests to process voice queries received by the system during a specified time interval satisfies one or more criteria. In response, the system triggers analysis of at least some of the requests received during the specified time interval to trigger analysis of at least some received requests to determine a set of requests that each identify a common voice query. The system can generate an electronic fingerprint that indicates a distinctive model of the common voice query. The fingerprint can then be used to detect an illegitimate voice query identified in a request from a client device at a later time.

Abstract: A system and method are presented for the correction of packet loss in audio in automatic speech recognition (ASR) systems. Packet loss correction, as presented herein, occurs at the recognition stage without modifying any of the acoustic models generated during training. The behavior of the ASR engine in the absence of packet loss is thus not altered. To accomplish this, the actual input signal may be rectified, the recognition scores may be normalized to account for signal errors, and a best-estimate method using information from previous frames and acoustic models may be used to replace the noisy signal.

Abstract: Methods and systems are provided for optimizing call quality and improving network efficiency by reducing bandwidth requirements at the individual-voice-call level. Embodiments provided herein build vocal fingerprints that correspond to the frequency range of the human voice, as well as the frequency range of the voice of individual users. The vocal fingerprints are used minimize and reduce the transmission of background noise and ambient sound captured using HD voice while retaining the frequency range of a user's voice in HD voice. This filtered HD voice frequency range is then transmitted to recipients over the network. The reduced frequency range lowers bandwidth usage and conserves network resources, all while optimizing the call quality for individual users.

Abstract: An object is to provide a technique which can provide a highly valid recognition result while preventing unnecessary processing. A voice recognition device includes first to third voice recognition units, and a control unit. When it is decided based on recognition results obtained by the first and second voice recognition units to cause the third voice recognition unit to recognize an input voice, the control unit causes the third voice recognition unit to recognize the input voice by using a dictionary including a candidate character string obtained by at least one of the first and second voice recognition units.

Abstract: Provided is an apparatus for bi-directional sign language/speech translation in real time and method that may automatically translate a sign into a speech or a speech into a sign in real time by separately performing an operation of recognizing a speech externally made through a microphone and outputting a sign corresponding to the speech, and an operation of recognizing a sign sensed through a camera and outputting a speech corresponding to the sign.

Abstract: Disclosed herein are systems, methods, and non-transitory computer-readable storage media for speech synthesis. A system practicing the method receives a set of ordered lists of speech units, for each respective speech unit in each ordered list in the set of ordered lists, constructs a sublist of speech units from a next ordered list which are suitable for concatenation, performs a cost analysis of paths through the set of ordered lists of speech units based on the sublist of speech units for each respective speech unit, and synthesizes speech using a lowest cost path of speech units through the set of ordered lists based on the cost analysis. The ordered lists can be ordered based on the respective pitch of each speech unit. In one embodiment, speech units which do not have an assigned pitch can be assigned a pitch.

Abstract: Systems and methods for utilizing incremental processing of portions of output data to limit the time required to provide a response to a user request are provided herein. In some embodiments, portions of the user request for information can be analyzed using techniques such as automatic speech recognition (ASR), speech-to-text (STT), and natural language understanding (NLU) to determine the overall topic of the user request. One the topic has been determined, portions of the anticipated audio output data can be synthesized independently instead of waiting for the complete response. The synthesized portions can then be provided to the electronic device in anticipation of being output through one or more speakers on the electronic device, which speeds up the time that the response can be provided to the user.

Abstract: There is provided a mechanism for synchronizing a plurality of dynamic language models residing in a plurality of devices associated with a single user, each device comprising a dynamic language model. The mechanism is configured to: receive text data representing text that has been input by a user into one or more of the plurality of devices; train at least one language model on the text data; and provide the at least one language model for synchronizing the devices. There is also provided a system comprising the mechanism and a plurality of devices, and a method for synchronizing a plurality of dynamic language models residing in a plurality of devices associated with a single user.

Abstract: The present invention relates to transposing signals in time and/or frequency and in particular to coding of audio signals. More particular, the present invention relates to high frequency reconstruction (HFR) methods including a frequency domain harmonic transposer. A method and system for generating a transposed output signal from an input signal using a transposition factor T is described. The system comprises an analysis window of length La, extracting a frame of the input signal, and an analysis transformation unit of order M transforming the samples into M complex coefficients. M is a function of the transposition factor T. The system further comprises a nonlinear processing unit altering the phase of the complex coefficients by using the transposition factor T, a synthesis transformation unit of order M transforming the altered coefficients into M altered samples, and a synthesis window of length Ls, generating a frame of the output signal.

Abstract: A method for speech enhancement in speech communication devices and more specifically in hearing aids for suppressing stationary and non-stationary background noise in the input speech signal signals is disclosed. The method uses spectral subtraction wherein the noise spectrum is updated using quantile-based estimation without voice activity detection and the quantile values are approximated by dynamic quantile tracking without involving large storage and sorting of past spectral samples. The technique permits use of a different quantile at each frequency bin for noise estimation without introducing processing overheads. The preferred embodiment uses analysis-modification-synthesis based on Fast Fourier transform (FFT) and it can be integrated with other FFT-based signal processing techniques used in the hearing aids and speech communication devices.

Abstract: A circuit to compensate for a lost audio frame, comprising: an identifier configured to identify a reference audio segment with a first length followed by the lost audio frame with a second length; a searcher coupled to the identifier and configured to search for a first audio segment similar to the reference audio segment in a cached audio segment followed by the reference audio segment by utilizing a cross-correlation search; the identifier further configured to identify a second audio segment subsequent to the first audio segment as a pre-compensated audio frame; an adjustor coupled to the identifier and configured to adjust an amplitude of the second audio segment based on a scale factor; and an output coupled to the adjustor to output the adjusted second audio segment as a compensated audio frame.

Abstract: An automatic speech recognition (“ASR”) system produces, for particular users, customized speech recognition results by using data regarding prior interactions of the users with the system. A portion of the ASR system (e.g., a neural-network-based language model) can be trained to produce an encoded representation of a user's interactions with the system based on, e.g., transcriptions of prior utterances made by the user. This user-specific encoded representation of interaction history is then used by the language model to customize ASR processing for the user.

Abstract: A method for speech modeling by an electronic device is described. The method includes obtaining a real-time noise reference based on a noisy speech signal. The method also includes obtaining a real-time noise dictionary based on the real-time noise reference. The method further includes obtaining a first speech dictionary and a second speech dictionary. The method additionally includes reducing residual noise based on the real-time noise dictionary and the first speech dictionary to produce a residual noise-suppressed speech signal at a first modeling stage. The method also includes generating a reconstructed speech signal based on the residual noise-suppressed speech signal and the second speech dictionary at a second modeling stage.

Abstract: Disclosed herein are systems, methods, and non-transitory computer-readable storage media for approximating relevant responses to a user query with voice-enabled search. A system practicing the method receives a word lattice generated by an automatic speech recognizer based on a user speech and a prosodic analysis of the user speech, generates a reweighted word lattice based on the word lattice and the prosodic analysis, approximates based on the reweighted word lattice one or more relevant responses to the query, and presents to a user the responses to the query. The prosodic analysis examines metalinguistic information of the user speech and can identify the most salient subject matter of the speech, assess how confident a speaker is in the content of his or her speech, and identify the attitude, mood, emotion, sentiment, etc. of the speaker. Other information not described in the content of the speech can also be used.

Abstract: A method and system for achieving emotional text to speech. The method includes: receiving text data; generating emotion tag for the text data by a rhythm piece; and achieving TTS to the text data corresponding to the emotion tag, where the emotion tags are expressed as a set of emotion vectors; where each emotion vector includes a plurality of emotion scores given based on a plurality of emotion categories. A system for the same includes: a text data receiving module; an emotion tag generating module; and a TTS module for achieving TTS, wherein the emotion tag is expressed as a set of emotion vectors; and wherein emotion vector includes a plurality of emotion scores given based on a plurality of emotion categories.

Abstract: An audio decoder for providing a decoded audio information includes a arithmetic decoder for providing a plurality of decoded spectral values on the basis of an arithmetically-encoded representation of the spectral values and a frequency-domain-to-time-domain converter for providing a time-domain audio representation using the decoded spectral values. The arithmetic decoder is configured to select a mapping rule describing a mapping of a code value onto a symbol code in dependence on a context state. The arithmetic decoder is configured to determine or modify the current context state in dependence on a plurality of previously-decoded spectral values. The arithmetic decoder is configured to detect a group of a plurality of previously-decoded spectral values, which fulfill, individually or taken together, a predetermined condition regarding their magnitudes, and to determine the current context state in dependence on a result of the detection. An audio encoder uses similar principles.

Abstract: A method comprising: receiving input signals for multiple channels; and parameterizing the received input signals into parameters defining multiple different object spectra and defining a distribution of the multiple different object spectra in the multiple channels.

Abstract: This application discloses a method implemented by an electronic device to detect a signature event (e.g., a baby cry event) associated with an audio feature (e.g., baby sound). The electronic device obtains a classifier model from a remote server. The classifier model is determined according to predetermined capabilities of the electronic device and ambient sound characteristics of the electronic device, and distinguishes the audio feature from a plurality of alternative features and ambient noises. When the electronic device obtains audio data, it splits the audio data to a plurality of sound components each associated with a respective frequency or frequency band and including a series of time windows. The electronic device further extracts a feature vector from the sound components, classifies the extracted feature vector to obtain a probability value according to the classifier model, and detects the signature event based on the probability value.