Abstract: An intelligent terminal includes unique features and integrated components to provide the ultimate dynamic viewing experience. The intelligent terminal is transportable and can function as a tablet, a laptop, or a desktop. The intelligent terminal is also configured to receive display data from an electronic device, via at least one processor and a network interface. The intelligent terminal includes a display screen that is configured to rotate about an axis that is normal to the display screen and is further configured to modify its display pixel density, brightness, clarity, color, either automatically or in response to a user input. The modified display data, based on the axial orientation of the screen, a display data attribute, the ambient conditions, the user's viewing parameters, and an electronic device attribute, is further configured to generate an image for display on the display screen that provides the most optimal and comfortable dynamic viewing experience.

Abstract: Systems and method for image generation in a gaze tracking display. A gaze tracking display system includes a graphics processor and display circuitry. The graphics processor is configured to perform foveated rendering of image data, and to output foveated image data. The display circuitry is coupled to the graphics processor. The display circuitry includes a display device and a display controller. The display device is configured to produce a viewable image. The display controller is configured to drive the display device. The display controller includes foveated data reconstruction circuitry configured to produce an image at a resolution of the display device based on the foveated image data received from the graphics processor.

Abstract: The invention is directed to a percussion system to be attached to a stringed instrument in a way that allows a player to drum the percussion system while plucking the strings of the stringed instrument using a single hand. The percussion system includes a percussion device including: a body and at least one static percussion surface located over the body, the percussion surface having a batter side. The percussion system further includes a connecting element for connecting the percussion device to a stringed instrument such that when the percussion system is connected to the stringed instrument the percussion surface is substantially perpendicular to the stringed instrument's front side and is directed towards strings of the stringed instrument.

Abstract: Tonal musical instruments configured for outdoor installation are disclosed. The tonal musical instrument may comprise a support post, wherein the support post is configured for attachment to an outdoor surface, at least one a non-vibrating guard, a metal plate tuned to produce at least one note on a musical scale when struck by a user, and one or more fasteners securing the at least one non-vibrating guard and the metal plate to the support post.

Abstract: A timing control method includes generating a timing designation signal according to one of a first generation mode for generating the timing designation signal which designates, based on a detection result of a first event in a performance of a music piece, a timing of a second event in the performance, and a second generation mode for generating the timing designation signal without using the detection result, and outputting a command signal for commanding an execution of the second event according to one of a first output mode for outputting the command signal in accordance with the timing designated by the timing designation signal, and a second output mode for outputting the command signal in accordance with a timing determined based on the music piece.

Abstract: A pickup system for an acoustic instrument having strings. The pickup system includes a mounting device configured to be coupled with an acoustic instrument, a sensing element secured to the mounting device and configured to convert mechanical vibrations produced by strings of the acoustic instrument into an electrical signal, and a secondary element secured to the mounting device and configured to reduce pickup of noise produced by sources other then the strings of the acoustic instrument.

Abstract: A method is presented for estimating and suppressing reverberation from a digital reverberant signal. A method for changing a first reverberation estimation according to another reverberation estimation is further provided. A method for controlling the reverberation suppression rate is also presented.

Abstract: The disclosure includes an acoustic processing network comprising a Digital Signal Processor (DSP) operating at a first frequency and a Real-Time Acoustic Processor (RAP) operating at a second frequency higher than the first frequency. The DSP receives a noise signal from at least one microphone. The DSP then generates a noise filter based on the noise signal. The RAP receives the noise signal from the microphone and the noise filter from the DSP. The RAP then generates an anti-noise signal based on the noise signal and the noise filter for use in Active Noise Cancellation (ANC).

Abstract: A device for active noise control inside a vehicle, the device comprising a processor (7610) configured to determine a noise wavefield within the vehicle based on noise signals captured by a microphone array (M1-M10); determine the position of the ears of a passenger (P1) based on information obtained from a head tracking sensor (HTU1); capture a noise field inside the vehicle based on information obtained by the microphone array; obtain a noise level at the ears of the passenger (P1) from the noise field; and determine an anti-noise field based on the noise level at the position of the ears of the passenger.

Abstract: The disclosure relates to an active vibration absorption system for absorbing the vibrations of a vibrating element. The vibration absorption system comprises a structure-borne sound exciter having a coupling element for coupling to the vibrating element and having an electrical coil for moving the coupling element by means of a coil current. The structure-borne sound exciter is designed to provide a measurement signal correlated with a detected induced voltage. Furthermore, the vibration absorption system comprises a control device, which is designed to identify a target current intensity of the coil current in accordance with the measurement signal and to control an actual current intensity of the coil current to the target current intensity. The target current intensity is designed to adjust the motion of the coupling element in such a way that the vibrations of the element are at least partly absorbed.

Abstract: A speech processing device of an embodiment includes a spectrum parameter calculation unit, a phase spectrum calculation unit, a group delay spectrum calculation unit, a band group delay parameter calculation unit, and a band group delay compensation parameter calculation unit. The spectrum parameter calculation unit calculates a spectrum parameter. The phase spectrum calculation unit calculates a first phase spectrum. The group delay spectrum calculation unit calculates a group delay spectrum from the first phase spectrum based on a frequency component of the first phase spectrum. The band group delay parameter calculation unit calculates a band group delay parameter in a predetermined frequency band from a group delay spectrum. The band group delay compensation parameter calculation unit calculates a band group delay compensation parameter to compensate a difference between a second phase spectrum reconstructed from the band group delay parameter and the first phase spectrum.

Abstract: Embodiments of the present disclosure provide a language recognition method and apparatus, a device and a computer storage medium. In an aspect, in the embodiments of the present disclosure, after the Nth speech segment included by the speech signal is received, language recognition is performed according to already-received previous N speech segments to obtain the score of each language in at least one language, N being 2, 3, 4 . . . ; therefore, if there exists a langue whose score reaches the designated threshold, the language whose score reaches the designated threshold is considered as the language matched with the speech signal. Therefore, the technical solutions according to embodiments of the present disclosure solve the problem in the prior art that the efficiency of language recognition is lower so that the language recognition cannot be applied to an application scenario in which the recognition result needs to be obtained quickly.

Abstract: A voice recognition method is provided that includes extracting a first speech from the sound collected with a microphone connected to a voice processing device, and calculating a recognition result for the first speech and the confidence level of the first speech. The method also includes performing a speech for a repetition request based on the calculated confidence level of the first speech, and extracting with the microphone a second speech obtained through the repetition request. The method further includes calculating a recognition result for the second speech and the confidence level of the second speech, and generating a recognition result from the recognition result for the first speech and the recognition result for the second speech, based on the confidence level of the calculated second speech.

Abstract: A method, a computer program product, and a computer system for mapping between a speech signal and a transcript of the speech signal. The computer system segments the speech signal to obtain one or more segmented speech signals and the transcript of the speech signal to obtain one or more segmented transcripts of the speech signal. The computer system generates estimated phone sequences and reference phone sequences, calculates costs of correspondences between the estimated phone sequences and the reference phone sequences, determines a series of the estimated phone sequences with a smallest cost, selects a partial series of the estimated phone sequences from the series of the estimated phone sequences, and generates mapping data which includes the partial series of the estimated phone sequences and a corresponding series of the reference phone sequences.

Abstract: A “Facet Recommender” creates conversational recommendations for facets of particular conversational topics, and optionally for things associated with those facets, from consumer reviews or other social media content. The Facet Recommender applies a machine-learned facet model and optional sentiment-model, to identify facets associated with spans or segments of the content and to determine neutral, positive, or negative consumer sentiment associated with those facets and, optionally, things associated with those facets. These facets are selected by the facet model from a list or set of manually defined or machine-learned facets for particular conversational topic types. The Facet Recommender then generates new conversational utterances (i.e., short neutral, positive or negative suggestions) about particular facets based on the sentiments associated with those facets. In various implementations, utterances are fit to one or more predefined conversational frameworks.

Abstract: A system and method for speech recognition is provided. Embodiments may include receiving an audio signal at a first deep neural network (“DNN”) associated with a computing device. Embodiments may further include receiving the audio signal at a second deep neural network (“DNN”) associated with a computing device, wherein the second deep neural network includes fewer parameters than the first deep neural network. Embodiments may also include determining whether to select an output from the first deep neural network or the second deep neural network and providing the selected output to a decoder with an overall objective of speeding up ASR.

Abstract: A method, computer program product, and computer system for transforming, by a computing device, a speech signal into a speech signal representation. A regression deep neural network may be trained with a cost function to minimize a mean squared error between actual values of the speech signal representation and estimated values of the speech signal representation, wherein the cost function may include one or more discriminative terms. Bandwidth of the speech signal may be extended by extending the speech signal representation of the speech signal using the regression deep neural network trained with the cost function that includes the one or more discriminative terms.

Abstract: A method and system are directed to autonomous neural network keyphrase detection and includes generating and using a multiple element state score vector by using neural network operations and without substantial use of a digital signal processor (DSP) to perform the keyphrase detection.

Abstract: Systems and methods for providing a dynamically configurable interface for structured note dictation input for one or more electronic health record (EHR) systems generally comprise a dictation user interface, a dictation management subsystem, and at least one EHR system having a plurality of electronic medical records. The dictation management subsystem generally comprises a computing device configured to moderate the recognized text for a particular EHR system and particular electronic health record.

Abstract: The present disclosure provides a speech recognition method and device. The method includes: receiving a speech signal; decoding the speech signal according to an acoustic model, a language model and a decoding network established in advance, and dynamically adding a blank unit in a decoding process to obtain an optimum decoding path with the added blank unit, in which the acoustic model is obtained based on connectionist temporal classification training, the acoustic model includes basic pronunciation units and the blank unit, and the decoding network includes a plurality of decoding paths consisting of the basic pronunciation units; and outputting the optimum decoding path as a recognition result of the speech signal.

Abstract: Systems and methods of determining phonetic relationships are provided. For instance data indicative of an input text phrase input by a user can be received. An audio output corresponding to a spoken rendering of the input text phrase can be determined. A text transcription of the audio output of the input text phrase can be determined. The text transcription can be a textual representation of the audio output. The text transcription can be compared against a plurality of test phrases to identify a match between the text transcription and at least one test phrase.

Abstract: Disclosed in various examples are methods, systems, and machine-readable mediums for providing improved computer implemented speech recognition by detecting and correcting speech recognition errors during a speech session. The system recognizes repeated speech commands from a user in a speech session that are similar or identical to each other. To correct these repeated errors, the system creates a customized language model that is then utilized by the language modeler to produce a refined prediction of the meaning of the repeated speech commands. The custom language model may comprise clusters of similar past predictions of speech commands from the speech session of the user.

Abstract: A system for extracting verifiable entities from a user-utterance received on an automated calling service is provided. The system may include a receiver configured to receive a user-utterance, a processor and a non-transitory computer-readable media comprising computer-executable instructions. The processor may be configured to execute the instructions which, canonicalize the user-utterance into a plurality of tokens, determine the number of tokens of the user-utterance, and generate, using a sliding-window protocol, a comprehensive number of n-gram sequences from the user-utterance. The processor may be configured to process a plurality of threads of execution that may include a series of actions executed on the n-gram sequences to identify and extract verified entities from the user-utterance.

Abstract: A computer-implemented method for analyzing content written on a board, on which a text and/or a drawing are made, is disclosed. The method includes obtaining content data including a series of images, which captures content being written on the board. The method also includes obtaining utterance data representing a series of utterances, which is associated with the series of the images. The method further includes extracting a section from the series of the utterances based on a change in topics and recognizing a content block for the section from the content data. The content block includes one or more content parts written during the section. The method includes further calculating evaluation value for the content block by using one or more utterances included in the section.

Abstract: Provided are an interactive question answering apparatus and method, which improve an unnatural question sentence and prevent the impossibility of a question in a state which does not know additional information, thereby providing a multimodal-based question answering method enabling a person to provide a question like conversing. Also, a question answering service is exposed to potential product purchasers of content such as an image, video, audio, etc., thereby providing an environment available to advertisement market.

Abstract: A topic providing device includes a candidate topic extractor, a provided topic determiner, a voice synthesizer, and a speaker. When a determination is made that a parent and child are conversing and that there is a need to provide a new topic to the parent and child, based on a conversation history database and a child activity database storing at least one activity name indicating an activity the child was engaged in for a first predetermined period of time, the candidate topic extractor extracts at least one candidate topic that corresponds to the at least one activity name in the child activity database and does not correspond to an activity name included in text data recorded in a first database. From the at least one candidate topic, the provided topic determiner selects one topic to provide to the parent and the child. The voice synthesizer generates voice data containing the one topic. The speaker outputs the voice data.

Abstract: An artificial intelligence voice interactive system may provide various services to a user in response to a voice command. The system may perform at least one of: receiving user speech from a user; transmitting the received user speech to the central server; receiving at least one of control signals and audio and video answers, which are generated based on an analysis and interpretation result of the user speech, from at least one of the central server, the internal service servers, and the external service servers; and outputting the received audio and video answers through at least one of speakers and a display coupled to the user interactive device and controlling at least one device according to the control signals.

Abstract: Disclosed is an electronic device includes a display, a communication circuit, a processor, and a memory. The memory stores instructions that cause the processor to receive a natural language input from a user, to transmit at least a portion of the natural language input to an external server, to display a first message based on the natural language input and a first indication indicating the user, to receive a first response from the external server, to display a second indication indicating a first engine and a second message based on the first response, to receive a second response from the external server, and to display a third indication indicating a second engine and a third message based on the second response. The first response is generated by the first engine based on the natural language input, and the second response is generated by the second engine based on the natural language input.

Abstract: Disclosed herein is a method for performing semi-supervised question answering. The method includes the steps of: (i) creating a graph based on input data, the input data comprising natural language text that includes seen questions, background text, and seen answers, and (ii) querying the graph in response to a new question. Querying the graph in response to a new question includes the steps of: projecting the question into the graph to a projected location; and for each node in a retrieval radius associated with the question, determining a set of answer weights based on the distances between retrieved nodes and the projected location and answer weights associated with retrieved nodes based on seen questions; and determining a best response corresponding to the seen answer associated with the highest answer weight from the set of answer weights determined for each retrieved node within the retrieval radius.

Abstract: A method and system of providing a portable voice-based control user interface for multiple types of appliances are disclosed. The method includes activating a built-in voice communication interface of a voice control apparatus; selecting a first target appliance to receive voice-based commands; receiving a first voice input; in accordance with a determination that the first target appliance is a first appliance of a first appliance type, processing the first voice input using a first NLP model corresponding to the first appliance type to obtain a first machine command, and sending the first machine command to the first appliance; and in accordance with a determination that the first target appliance is a second appliance of a second appliance type, processing the first voice input using a second NLP model corresponding to the second appliance type to obtain a second machine command, and sending the second machine command to the second appliance.

Abstract: Methods, apparatus, and computer readable media related to receiving textual input of a user during a dialog between the user and an automated assistant (and optionally one or more additional users), and generating responsive reply content based on the textual input and based on user state information. The reply content is provided for inclusion in the dialog. In some implementations, the reply content is provided as a reply, by the automated assistant, to the user's textual input and may optionally be automatically incorporated in the dialog between the user and the automated assistant. In some implementations, the reply content is suggested by the automated assistant for inclusion in the dialog and is only included in the dialog in response to further user interface input.

Abstract: Implementations relate to executing a tailored version of a dynamic interactive dialog application, where the tailored version is tailored based on structured content that is specified by a creator of the tailored version. Executing the tailored version of the interactive dialog application can be in response to receiving, via an assistant interface of an assistant application, an invocation phrase assigned to the tailored version and/or other user interface input that identifies the tailored version. In some implementations, a tailored version of a dynamic interactive dialog application is executed with persona value(s) that are specified by a creator of the tailored version and/or that are predicted based on structured content and/or other input provided by the creator in creating the tailored version. In some implementations, structured content and/or other input provided by a creator in creating a tailored version of an interactive dialog application is utilized in indexing the tailored version.

Abstract: A server controlling an external device is provided. The server includes a communicator; a processor; a memory which stores at least one natural language understanding (NLU) engine for generating a command corresponding to a user's utterance. The server receives, from a pairing device paired to the external device, the user's utterance controlling the external device and information about at least one external device registered with the pairing device, via the communicator, determines an NLU engine corresponding to the external device, from among the at least one NLU engine, based on the user's utterance controlling the external device and the information about the at least one external device, and generates the command controlling the external device based on the user's utterance, by using the determined NLU engine.

Abstract: A voice-enabled digital communications assistant powered in part using tailored machine learning models and other algorithms is used to engage with and control one or more healthcare devices or instruments such that a user is able to control the devices or instruments using natural language, conversational-like, voice commands. A command processor processes the audible instructions, while a context-aware processor monitors the present states and conditions of all devices and instruments, as well as the environment, for situational awareness purposes, including situations where executing commands may be incompatible with or conflict with the present states or conditions of devices and instruments as well as their expected future states. In addition to speech responses by the digital assistant, a separate notification engine provides audible or visual feedback to the user.

Abstract: A computerized method of authenticating a user includes receiving, by a computing device, first biometric data from the user via a first biometric input device; extracting one or more first features from the first biometric data; generating a first vector based on the first features; computing a first similarity score reflecting a geometric distance between the first vector and a first reference vector stored in memory; receiving second biometric data from the user via a second biometric input device; extracting one or more second features from the second biometric data; generating a second vector based on the second features; computing a second similarity score reflecting a geometric distance between the second vector and a second reference vector stored in memory; computing a confidence score based on the first and second similarity scores; and determining a right of the user to access restricted information based on the confidence score.

Abstract: Provided is a technology which improves reliability of the interaction between devices in a system where the devices communicate. In an information appliance system, multiple information appliances and a communication device such as a smart phone are in M2M communication. The communication device receives input of a voice from a user, and authenticates the user based on a voice signal, and sample voice data accumulated for user identification. The communication device performs a speech recognition process on the voice signal to determine an instruction of the user. When the user is authenticated and the instruction of the user is determined, the communication device transmits a command according to the instruction to an information appliance that is associated with the determined instruction of the user.

Abstract: Systems and method of diarization of audio files use an acoustic voiceprint model. A plurality of audio files are analyzed to arrive at an acoustic voiceprint model associated to an identified speaker. Metadata associate with an audio file is used to select an acoustic voiceprint model. The selected acoustic voiceprint model is applied in a diarization to identify audio data of the identified speaker.

Abstract: Provided is a method including: receiving an audio signal of a transmitter; detecting sensitive information in the audio signal based on content of the audio signal; encrypting the sensitive information by using characteristic information of a receiver; and transmitting the audio signal including the encrypted sensitive information.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving audio data corresponding to an utterance, determining that the audio data corresponds to a hotword, generating a hotword audio fingerprint of the audio data that is determined to correspond to the hotword, comparing the hotword audio fingerprint to one or more stored audio fingerprints of audio data that was previously determined to correspond to the hotword, detecting whether the hotword audio fingerprint matches a stored audio fingerprint of audio data that was previously determined to correspond to the hotword based on whether the comparison indicates a similarity between the hotword audio fingerprint and one of the one or more stored audio fingerprints that satisfies a predetermined threshold, and in response to detecting that the hotword audio fingerprint matches a stored audio fingerprint, disabling access to a computing device into which the utterance was spoken.

Abstract: Methods, systems and computer program products for operating a voice response system in a multiuser environment are provided. Aspects include receiving a voice command from a first user and determining an identity of the first user based at least in part on a voice recognition of the first user. Aspects also include determining an identity of one or more other users in range of the voice response system and obtaining a command hierarchy. Aspects further include performing an action requested by the voice command based on a determination that the first user is authorized to request the voice command, wherein the determination that the first user is authorized to request the voice command is based at least upon the identity of the first user, the identity of one or more other users in range of the voice response system and the command hierarchy.

Abstract: Processing circuitry of an information processing apparatus obtains a set of identity vectors that are calculated according to voice samples from speakers. The identity vectors are classified into speaker classes respectively corresponding to the speakers. The processing circuitry selects, from the identity vectors, first subsets of interclass neighboring identity vectors respectively corresponding to the identity vectors and second subsets of intraclass neighboring identity vectors respectively corresponding to the identity vectors. The processing circuitry determines an interclass difference based on the first subsets of interclass neighboring identity vectors and the corresponding identity vectors; and determines an intraclass difference based on the second subsets of intraclass neighboring identify vectors and the corresponding identity vectors.

Abstract: Various embodiments of the technology described herein alleviate the need to specifically request enrollment information from a user to enroll the user in a voice biometric authentication program. For example, the system can receive a voice interaction from a user where the voice interaction includes a request or a command having one or more portions. After the user is authenticated using non-voice biometric authentication information, the system enrolls the user into a voice biometric authentication program for at least one portion of the request or the command. The system also enrolls the user into the voice biometric authentication program for other requests or commands that include one of the at least one portion of the request or the command and portions of one or more second requests or commands for which the user has been enrolled into the voice biometric authentication program.

Abstract: An information processing device including a connection control unit that, in a case where information regarding a situation of a user is detected, controls connection to another information processing device by using identification information corresponding to the situation of the user, an output control unit configured to control output of acquisition information acquired from the other information processing device, while the information processing device is connected to the other information processing device, and a transmission control unit configured to control transmission of input information input by the user, to the other information processing device, while the information processing device is connected to the other information processing device. In a case where a predetermined condition is satisfied, the connection control unit invalidates the identification information.

Abstract: A method for generating a bitstream indicative of an object based audio program is described. The bitstream comprises a sequence of containers. A first container of the sequence of containers comprises a plurality of substream entities for a plurality of substreams of the object based audio program and a presentation section. The method comprises determining a set of object channels. The method further comprises providing a set of object related metadata for the set of object channels. In addition, the method comprises inserting a first set of object channel frames and a first set of object related metadata frames into a respective set of substream entities of the first container. Furthermore, the method comprises inserting presentation data into the presentation section.

Abstract: An audio processing method includes the following operation: dividing an audio file into a plurality of audio segments, in which a processing of a first audio segment of the audio segments includes the following operations: analyzing a first lowest energy value in a spectrum of the first audio segment; comparing the first minimum energy value with a preset energy value, and using a higher one as a first noise floor; generating a first processed audio segment according to the first noise floor and the first audio segment; compressing the first processed audio segment to produce a compressed audio segment; and sending the compressed audio segment to an audio playback device.

Abstract: The following coding scenario is addressed: A number of audio source signals need to be transmitted or stored for the purpose of mixing wave field synthesis, multi-channel surround, or stereo signals after decoding the source signals. The proposed technique offers significant coding gain when jointly coding the source signals, compared to separately coding them, even when no redundancy is present between the source signals. This is possible by considering statistical properties of the source signals, the properties of mixing techniques, and spatial hearing. The sum of the source signals is transmitted plus the statistical properties of the source signals which mostly determine the perceptually important spatial cues of the final mixed audio channels. Source signals are recovered at the receiver such that their statistical properties approximate the corresponding properties of the original source signals. Subjective evaluations indicate that high audio quality is achieved by the proposed scheme.

Abstract: Example embodiments disclosed herein relate to signal processing. A method for decomposing a plurality of audio signals from at least two different channels is disclosed. The method comprises obtaining a set of components that are weakly correlated, the set of components generated based on the plurality of audio signals. The method comprises extracting a feature from the set of components, and determining a set of gains associated with the set of components at least in part based on the extracted feature, each of the gains indicating a proportion of a diffuse part in the associated component. The method further comprises decomposing the plurality of audio signals by applying the set of gains to the set of components. Corresponding system and computer program product are also disclosed.

Abstract: Network communication speech handling systems are provided herein. In one example, a method of processing audio signals by a network communications handling node is provided. The method includes processing an audio signal to determine a pitch cycle property associated with the audio signal, determining transfer times for encoded segments of the audio signal based at least in part on the pitch cycle property, and transferring packets comprising one or more encoded segments for delivery to a target node in accordance with the transfer time.

Abstract: Techniques are provided for reduction of echo in a received audio signal based on infinite impulse response (IIR) acoustic echo cancellation in the frequency domain. A methodology implementing the techniques according to an embodiment includes estimating an echo path transfer function associated with the received audio signal. The received audio signal includes a combination of a speech signal and a reference signal modified by the echo path transfer function. The estimation employs an IIR filter and a finite impulse response (FIR) filter, both of which operate in the frequency domain. The IIR filter is configured to model longer term echo components and the FIR filter is configured to model shorter term echo components. The method further includes applying the filters to the reference signal to generate an echo correction signal which is subtracted from the received audio signal to reduce the echo and generate an estimate of the speech signal.

Abstract: A device that determines an echo latency estimate by subsampling reference audio data. The device may determine the echo latency corresponding to an amount of time between sending reference audio data to loudspeaker(s) and microphone audio data corresponding to the reference audio data being received. The device may generate subsampled reference audio data by selecting only portions of the reference audio data that have a magnitude above a desired percentile. For example, the device may compare a magnitude of an individual reference audio sample to a percentile estimate value and sample only the reference audio samples that exceed the percentile estimate value. The device generate cross-correlation data between the subsampled reference audio data and the microphone audio data and may estimate the echo latency based on an earliest significant peak represented in the cross-correlation data.