Abstract: A method for reconstructing at least one target signal comprises determining a first set of feature vectors from the input signal, the first set of feature vectors forming a non-negative input matrix; determining a second set of feature vectors, the second set of feature vectors forming a non-negative noise matrix; decomposing the input matrix into a sum of a first matrix and a second matrix, the first matrix representing a product of a non-negative bases matrix and a non-negative weight matrix, and the second matrix representing a combination of the noise matrix and a noise weight vector; and reconstructing the at least one target signal based on the non-negative bases matrix and the non-negative weight matrix.

Abstract: A speech signal is received at an input. At least one electrical value associated with the received speech signal is tracked. A dynamic adjustment of the speech signal is determined. The dynamic adjustment is selected at least in part so as to minimize a distortion and minimize an over-amplification of the speech signal based at least in part upon an analysis of the at least one electrical value. The dynamic adjustment is further selected to obtain a desired output signal characteristic for the speech signal presented at an output. The dynamic adjustment value is applied to the speech signal and the adjusted speech signal is presented at the output. The gain of the signal can also be limited to prevent over-amplification.

Abstract: A method for translation supply chain analytics includes receiving operational variables of a translation process from a translation supply chain. The method further includes determining a cognitive leverage and a productivity factor for post editing of matches of a plurality of match types generated by the translation supply chain based at least in part on the operational variables from the translation supply chain. The method further includes generating linguistic markers for the matches of the plurality of match types generated by the translation supply chain, based at least in part on the cognitive leverage and the productivity factor for the post editing of the matches of the plurality of match types. The method further includes performing statistical analysis of the linguistic markers for the matches of the plurality of match types. The method further includes generating one or more analytics outputs based on the statistical analysis of the linguistic markers.

Abstract: To obtain a high-quality enhanced signal, disclosed is a signal processing apparatus including a transform unit that transforms a mixed signal in which a first signal and a second signal coexist, into a phase component and a magnitude component or power component for each frequency, a first control unit that replaces the phase component of a predetermined frequency, a second control unit that modifies the magnitude component or power component of the predetermined frequency in accordance with the amount of a change of the magnitude component or power component that arises from replacement by the first control unit, and a reconstruction unit that reconstructs the phase component replaced by the first control unit and the magnitude component or power component modified by the second control unit.

Abstract: A low overhead voice activity detection technique for a noise-canceling bioacoustic sensor consumes, as inputs, signals generated by a body microphone and an environment microphone and produces, as an output based on these inputs, an indication of whether voice activity is present. The technique applies a novel dual ANC configuration that produces, in addition to the normal noise reduction function, a signal composed of environmental sounds with body sounds attenuated and a signal composed of body sounds projected into the acoustic environment, primarily speech sounds. The technique then applies to these derived signals an algorithm based on the scientific observation that speech intensity, within limits, rises and falls to match environmental sound intensity to provide voice activity detection.

Abstract: A technique for enhancing speech signal captured in a noisy environment is provided. According an example embodiment, the technique comprises obtaining a current time frame of a noise-suppressed voice signal, derived on basis of a current time frame of a source audio signal comprising a source voice signal, detecting input voice characteristics for the current time frame of noise-suppressed voice signal, obtaining reference voice characteristics for said current time frame, said reference voice characteristics being descriptive of the source voice signal in noise-free or low-noise environment, and creating a current time frame of a modified voice signal by modifying said current time frame of the noise-suppressed voice signal in response to a difference between the detected input voice characteristic and the reference voice characteristics exceeding a predetermined threshold.

Abstract: Embodiments included herein are directed towards a system and method for compressed domain language identification. Embodiments may include receiving a bitstream of a sequence of packets at one or more computing devices and classifying each packet into speech or non-speech based upon, at least in part, compressed domain voice activity detection (VAD). Embodiments may further include extracting a pseudo-cepstral representation from the speech detected packets and partially decoding without extracting a PCM format and generating a sequence of multi-frames, based upon, at least in part, the pseudo-cepstral representation. Embodiments may also include providing in real time the sequence of multi-frames to a deep neural network (DNN), wherein the DNN has been trained off-line for one or more desired target languages.

Abstract: Provided are an apparatus and method for reporting speech recognition failures. The method includes detecting pure speech data from input speech data and outputting the detected pure speech data, determining at least one speech recognition failure for the pure speech data, and ascertaining speech recognition failure reasons from a check result for the speech recognition failures and outputting the ascertained speech recognition failure reasons.

Abstract: Techniques related to implementing neural networks for speech recognition systems are discussed. Such techniques may include implementing frame skipping with approximated skip frames and/or distances on demand such that only those outputs needed by a speech decoder are provided via the neural network or approximation techniques.

Abstract: Described herein is a speech recognition device comprising: a communication module receiving speech data corresponding to speech input from a speech recognition terminal and multi-sensor data corresponding to input environment of the speech; a model selection module selecting a language and acoustic model corresponding to the multi-sensor data among a plurality of language and acoustic models classified according to the speech input environment on the basis of previous multi-sensor data; and a speech recognition module controlling the communication module to apply a feature vector extracted from the speech data to the language and acoustic model and transmit speech recognition result for the speech data to the speech recognition terminal.

Abstract: A method includes receiving an input signal comprising an original domain signal and creating a first window data set and a second window data set from the signal, wherein an initiation of the second window data set is offset from an initiation of the first window data set, converting the first window data set and the second window data set to a frequency domain and storing the resulting data as data in a second domain different from the original domain, performing complex spectral phase evolution (CSPE) on the second domain data to estimate component frequencies of the first and second window data sets, using the component frequencies estimated in the CSPE, sampling a set of second-domain high resolution windows to select a mathematical representation comprising a second-domain high resolution window that fits at least one of the amplitude, phase, amplitude modulation and frequency modulation of a component of an underlying signal wherein the component comprises at least one oscillator peak, generating an ou

Abstract: In accordance with embodiments of the present disclosure, an adjustable equalization filter may have a response that generates an equalized source audio signal from a source audio signal to account for effects of changes in an electro-acoustical path of the source audio signal to a transducer. An equalizer coefficient control block may adapt the response of the adjustable equalization filter in response to changes in a response of a secondary path estimate filter for modeling the electro-acoustical path of a source audio signal through the transducer, wherein a response of the secondary path estimate filter is adapted in conformity with an error microphone signal indicative of the acoustic output of the transducer.

Abstract: A system can receive a gesture from a user and configure a microphone system based on the received gesture to be more sensitive in the direction of a user from a device. The gesture can detected by a sensor and can be a touch input, input from a camera and a depth sensor and the like. The microphone system can include a microphone that can be electronically or mechanically steerable, or both. Acoustic signals received from the direction of the user and from other directions can be used in conjunction with an automatic speech recognition system to detect and process a command from the user.

Abstract: Compensating for identifiable background content in a speech recognition device, including: receiving, by a noise filtering module, an identification of environmental audio data received by the speech recognition device; and filtering, by the noise filtering module in dependence upon which portion of the identified environmental audio data was being rendered when the audio data generated from the plurality of sources was received, the audio data generated from the plurality of sources.

Abstract: A method and system can control a speech recognition system in a vehicle. The method includes monitoring adaptive feature data about interactions between a user and the speech recognition system. The method includes determining a first group of samples of the adaptive feature data and creating a control chart based on the first group of samples. The control chart includes a control limit. The method further includes determining a second group of samples of the adaptive feature data after creating the control chart. Furthermore, the method includes calculating an arithmetic mean of each sample of the second group of samples to determine a sample mean, comparing the sample mean to the control limit in order identify unexpected performance of the speech recognition system. The method includes adjusting the speech recognition system based on the identified unexpected performance if the unexpected performance is identified.

Abstract: A method of processing a signal, including taking a signal formed from a plurality of source signal emitters and expressed in an original domain, decomposing the signal into a mathematical representation of a plurality of constituent elements in an alternate domain, analyzing the plurality of constituent elements to associate at least a subset of the constituent elements with at least one of the plurality of source signal emitters, separating at least a subset of the constituent elements based on the association and reconstituting at least a subset of constituent elements to produce an output signal in at least one of the original domain, the alternate domain and another domain.

Abstract: A wireless headset capable of receiving audio signals transmitted wirelessly and compatible for use in an MRI scanner is disclosed. The headset includes a first wireless module connected to the first earphone and a second wireless module connected to the second earphone. Each wireless module is electrically connected to a speaker in the respective earphone. The first wireless module receives the audio signal from a remote source and coordinates transmission of the audio signal to each of the speakers. The compact nature of each earphone minimizes the length of wire runs. In addition, the headset is made of materials having low magnetic susceptibility such that they will not be affected by the magnetic field from the MRI scanner.

Abstract: Systems and methods for reducing noise from an input signal are provided. An input signal is received. The input signal is transformed from a time domain to a plurality of subbands in a frequency domain, where each subband of the plurality of subbands includes a speech component and a noise component. For each of the subbands, an amplitude of the speech component is estimated based on an amplitude of the subband and an estimate of at least one signal-to-noise ratio (SNR) of the subband. The estimating of the amplitude of the speech component is based on a closed-form solution. The plurality of subbands in the frequency domain are filtered based on the amplitudes of the speech components.

Abstract: Sound decomposition models are described. In one or more implementations, a plurality of individual models is generated for respective ones of a plurality of sound sources. The plurality of models is collected to form a universal audio model that is configured to support sound decomposition of sound data through use of one or more of the models. The plurality of models is not generated using a sound source that originated at least a portion of the sound data.

Abstract: Speech from a driver and speech from a passenger in a vehicle is selected directionally using two or more microphones. Samples of speech from a driver and picked up by a first microphone are delayed until samples of the speech picked up by a second microphone is in phase with the speech picked up by the first microphone. Samples of a passenger's speech and picked up by the second microphone are delayed until samples of the passenger's speech picked up by the first microphone are out-of-phase with the speech picked up by the second microphone.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for hotword detection on multiple devices are disclosed. In one aspect, a method includes the actions of receiving, by a computing device, audio data that corresponds to an utterance. The actions further include determining a likelihood that the utterance includes a hotword. The actions further include determining a loudness score for the audio data. The actions further include based on the loudness score, determining an amount of delay time. The actions further include, after the amount of delay time has elapsed, transmitting a signal that indicates that the computing device will initiate speech recognition processing on the audio data.

Abstract: According to one aspect, a method for detecting voice activity is disclosed, the method including receiving a frame of an input audio signal, the input audio signal having an sample rate; dividing the frame into a plurality of subbands based on the sample rate, the plurality of subbands including at least a lowest subband and a highest subband; filtering the lowest subband with a moving average filter to reduce an energy of the lowest subband; estimating a noise level for each of the plurality of subbands; calculating a signal to noise ratio value for each of the plurality of subbands; and determining a speech activity level of the frame based on an average of the calculated signal to noise ratio values and a weighted average of an energy of each of the plurality of subbands. Other aspects include audio decoders that decode audio that was encoded using the methods described herein.

Abstract: Sound related vehicle information representing one or more sounds may be received in a processor associated with a vehicle. The sound related vehicle information may or may not include an audio signal. An audio signal output to a passenger may be modified based on the sound related vehicle information.

Abstract: A content exchange server facilitates the identification of potential narrators of content. The content exchange server receives audio samples from narrators and extracts recording features of the narrators from the samples, which can be associated with narrator profiles. A rights holder can submit preferred recording features for a work to the content exchange server. The content exchange server can compare the preferred focal features with the recording features extracted from the voice samples of the potential narrators to identify potential narrators for the work.

Abstract: An apparatus for adjusting a microphone sampling rate, the apparatus including an input to receive an audio signal from a microphone and a front-end processing module. The front-end processing module is to generate a plurality of frames from the audio signal received by the microphone, determine a noise profile using the plurality of frames, and adjust a sampling rate of the microphone based on the determined noise profile.

Abstract: Disclosed are an apparatus and a method for detecting a speech endpoint using a WFST. The apparatus in accordance with an embodiment of the present invention includes: a speech decision portion configured to receive frame units of feature vector converted from a speech signal and to analyze and classify the received feature vector into a speech class or a noise class; a frame level WFST configured to receive the speech class and the noise class and to convert the speech class and the noise class to a WFST format; a speech level WFST configured to detect a speech endpoint by analyzing a relationship between the speech class and noise class and a preset state; a WFST combination portion configured to combine the frame level WFST with the speech level WFST; and an optimization portion configured to optimize the combined WFST having the frame level WFST and the speech level WFST combined therein to have a minimum route.

Abstract: Methods, systems, and products for testing a grammar used in speech recognition for reliability in a plurality of operating environments having different background noise that include: receiving recorded background noise for each of the plurality of operating environments; generating a test speech utterance for recognition by a speech recognition engine using a grammar; mixing the test speech utterance with each recorded background noise, resulting in a plurality of mixed test speech utterances, each mixed test speech utterance having different background noise; performing, for each of the mixed test speech utterances, speech recognition using the grammar and the mixed test speech utterance, resulting in speech recognition results for each of the mixed test speech utterances; and evaluating, for each recorded background noise, speech recognition reliability of the grammar in dependence upon the speech recognition results for the mixed test speech utterance having that recorded background noise.

Abstract: The present invention relates to a listening device for a hearing impaired person. The present invention furthermore relates to a corresponding operating method of operating a listening device and to a corresponding computer program. In particular, the present invention relates to a listening device that comprises a signal processing unit that is controlled by a controller configured to implement a combined feed-forward and feed-back control in order to ensure that both an electric input signal and a processed electric output signal have at least almost identical modulation index values. Thereby, speech intelligibility is increased, in particular for a hearing impaired person being capable of perceiving sound pressure levels in a substantially decreased dynamic range.

Abstract: A sound processing device includes a first noise suppression unit configured to suppress a noise component included in an input sound signal using a first suppression amount, a second noise suppression unit configured to suppress the noise component included in the input sound signal using a second suppression amount greater than the first suppression amount, a speech section detection unit configured to detect whether the sound signal whose noise component has been suppressed by the second noise suppression unit includes a speech section having a speech for every predetermined time, and a speech recognition unit configured to perform a speech recognizing process on a section, which is detected to be a speech section by the speech section detection unit, in the sound signal whose noise component has been suppressed by the first noise suppression unit.

Abstract: An audio signal with a temporal sequence of blocks or frames is received or accessed. Features are determined as characterizing aggregately the sequential audio blocks/frames that have been processed recently, relative to current time. The feature determination exceeds a specificity criterion and is delayed, relative to the recently processed audio blocks/frames. Voice activity indication is detected in the audio signal. VAD is based on a decision that exceeds a preset sensitivity threshold and is computed over a brief time period, relative to blocks/frames duration, and relates to current block/frame features. The VAD and the recent feature determination are combined with state related information, which is based on a history of previous feature determinations that are compiled from multiple features, determined over a time prior to the recent feature determination time period. Decisions to commence or terminate the audio signal, or related gains, are outputted based on the combination.

Abstract: Apparatus, computer-readable storage medium, and method associated with orienting a display image are described. In embodiments, a computing device may include a display to render the display image and a display orientation module coupled with the display. In embodiments the display orientation module may receive audio input from a user of the computing device and determine a position of the user relative to the display, based on the audio input. In embodiments, the display orientation module may further either orient the display image in accordance with the position of the user or output a result of the determination for use to orient the display image in accordance with the position of the user. Other embodiments may be described and/or claimed.

Abstract: A system and method for reducing or entirely canceling background or ambient noise from a voice transmission from a communications device. A communications device, such as a mobile telephone, is configured with an ambient noise compensation signal generator that is connected between a microphone and a mixer. The original output of the microphone and a compensation signal generated by the ambient noise compensation signal generator are mixed together prior to being passed to a transmitter. In one embodiment a buffer is provided between the microphone and the mixer to help synchronize the timing of the signals to be mixed. In another embodiment a second microphone is employed to detect ambient noise.

Abstract: Various embodiments provide an ability to analyze an audio input signal and generate a counter audio signal based, at least in part, on the audio input signal. In some cases, combining the audio input signal with the counter audio signal renders the audio input signal incoherent and/or unintelligible to accidental listeners and/or listeners to whom the audio input signal is not directed towards. Alternately or additionally, the counter signal can mask the audio input signal to the accidental listeners.

Abstract: Disclosed are systems, methods, and computer readable media for identifying an acoustic environment of a caller. The method embodiment comprises analyzing acoustic features of a received audio signal from a caller, receiving meta-data information based on a previously recorded time and speed of the caller, classifying a background environment of the caller based on the analyzed acoustic features and the meta-data, selecting an acoustic model matched to the classified background environment from a plurality of acoustic models, and performing speech recognition as the received audio signal using the selected acoustic model.

Abstract: A method includes determining a first feature of a first audio signal at a first location in a signal processing path and determining, using the first feature, a first environmental classification of the first signal. Further, the method includes, based on the first environmental classification, enabling, modifying or disabling one or both of a first signal processing mode at the first location and a second signal processing mode at a second location in the signal processing path. The method also includes determining a second feature of a second audio signal at the second location and determining, using the second feature, a second environmental classification of the second signal. Further, the method includes, based on the second environmental classification, enabling, modifying or disabling one or both of the first signal processing mode at the first location and the second signal processing mode at the second location.

Abstract: A sound signal processing apparatus includes an observed signal analysis unit that receives as an observed signal a sound signal for channels obtained by a sound signal input unit formed of microphones and estimates a sound direction and a sound segment of a target sound which is sound to be extracted and a sound source extraction unit that receives the sound direction and sound segment of the target sound estimated by the observed signal analysis unit and extracts the sound signal for the target sound. The observed signal analysis unit includes a short time Fourier transform unit that generates an observed signal in time-frequency domain by applying short time Fourier transform to the sound signal for the channels received and a direction/segment estimation unit that receives the observed signal generated by the short time Fourier transform unit and detects the sound direction and sound segment of the target sound.

Abstract: Disclosed herein are systems, methods, and non-transitory computer-readable storage media for approximating an accent source. A system practicing the method collects data associated with customer specific services, generates country-specific or dialect-specific weights for each service in the customer specific services list, generates a summary weight based on an aggregation of the country-specific or dialect-specific weights, and sets an interactive voice response system language model based on the summary weight and the country-specific or dialect-specific weights. The interactive voice response system can also change the user interface based on the interactive voice response system language model. The interactive voice response system can tune a voice recognition algorithm based on the summary weight and the country-specific weights. The interactive voice response system can adjust phoneme matching in the language model based on a possibility that the speaker is using other languages.

Abstract: This method comprises steps of: a) partitioning (10, 16) the spectrum of the noisy signal into a HF part and a LF part; b) operating denoising processes in a differentiated manner for each of the two parts of the spectrum with, for the HF part, a denoising by prediction of the useful signal from one sensor to the other between sensors of a first sub-array (R1), by means of a first adaptive algorithm estimator (14), and, for the LF part, a denoising by prediction of the noise from one sensor to the other between sensors of a second sub-array (R2), by means of a second adaptive algorithm estimator (18); c) reconstructing the spectrum by combining together (22) the signals delivered after denoising of the two parts of the spectrum, respectively; and d) selectively reducing the noise (24) by an Optimized Modified Log-Spectral Amplitude gain, OM-LSA, process.

Abstract: A network infrastructure device of an apparatus in one example comprises a receive interface and a transmit interface for a communication channel. The network infrastructure device is configured to pass a data stream for the communication channel from the receive interface, through a set of signal processing blocks, and to the transmit interface. The network infrastructure device is configured to analyze the data stream for an occurrence of a predetermined data pattern. The network infrastructure device is configured to automatically add and/or remove one or more signal processing blocks from the set of signal processing blocks upon the occurrence of the predetermined data pattern.

Abstract: Methods and apparatuses for context assisted noise reduction are disclosed. In one example, noise data associated with background noise detected by a microphone at a mobile device is received. The noise data is processed to identify whether a threshold noise level has been exceeded. An event notification is transmitted, where the event notification is operable to initiate identifying a location having a reduced background noise.

Abstract: A method of detecting a user's voice activity in a mobile device is described herein. The method starts with a voice activity detector (VAD) generating a VAD output based on (i) acoustic signals received from microphones included in the mobile device and (ii) data output by an inertial sensor that is included in an earphone portion of the mobile device. The inertial sensor may detect vibration of the user's vocal chords modulated by the user's vocal tract based on vibrations in bones and tissue of the user's head. A noise suppressor may then receive the acoustic signals from the microphones and the VAD output and suppress the noise included in the acoustic signals received from the microphones based on the VAD output. The method may also include steering one or more beamformers based on the VAD output. Other embodiments are also described.

Abstract: A method of determining contexts for a vehicle, each context corresponding to one or more events associated with the vehicle, for example that the radio is on and a window is open. The method comprises detecting sound activities in an audio signal captured in the vehicle, and assigning context to the vehicle based on the detected sound activities. Non-audio data such as the operational status of a vehicle system or device is used to help assign contexts.

Abstract: A method for signal level matching by an electronic device is described. The method includes capturing a plurality of audio signals from a plurality of microphones. The method also includes determining a difference signal based on an inter-microphone subtraction. The difference signal includes multiple harmonics. The method also includes determining whether a harmonicity of the difference signal exceeds a harmonicity threshold. The method also includes preserving the harmonics to determine an envelope. The method further applies the envelope to a noise-suppressed signal.

Abstract: A first communication device receives a communication from a second communication device. The communication could be part of the signaling used to establish a call to the second communication device or an audio signal that the second communication device conveys after its user has answered the call and started speaking. Based on the communication from the second communication device, the first communication device provides a user-discernible indication regarding the environment in which the second communication device is being used, such as an indication that the second communication device is being used with a speakerphone. If the second communication device is being used in a vehicle, the first communication device may also provide a user-discernible indication regarding the operational status of the vehicle, such as whether the vehicle is in motion and/or the number of occupants in the vehicle.

Abstract: Audio frames are classified as either speech, non-transient background noise, or transient noise events. Probabilities of speech or transient noise event, or other metrics may be calculated to indicate confidence in classification. Frames classified as speech or noise events are not used in updating models (e.g., spectral subtraction noise estimates, silence model, background energy estimates, signal-to-noise ratio) of non-transient background noise. Frame classification affects acceptance/rejection of recognition hypothesis. Classifications and other audio related information may be determined by circuitry in a headset, and sent (e.g., wirelessly) to a separate processor-based recognition device.

Abstract: Provided is a noise-robust voice activity segmentation device which updates parameters used in the determination of voice-active segments without burdening the user, and also provided are a voice activity segmentation method and a voice activity segmentation program.

Abstract: Mitigating disruption to a voice recognition system in a vehicle caused by a passing source of noise is provided. Sensors sense an approaching truck or the like that is likely to disrupt operation of the in-vehicle voice recognition system. Countermeasures are initiated to mitigate the disruption.

Abstract: A method for processing a plurality of audio streams at a computer system onboard a vehicle is provided. The method receives the plurality of audio streams from a plurality of locations within a vehicle; prioritizes each of the plurality of audio streams to obtain a prioritization result; and completes a task associated with each of the plurality of audio streams, according to the prioritization result.

Abstract: A method for noise reduction is provided including: beamforming audio signals sampled by a microphone array to get a signal with an enhanced target voice and a signal with a weakened target voice; locating a target voice in the audio signal sampled by the microphone array; determining a credibility of the target voice when the target voice is located; updating an adaptive filter coefficient according to the credibility, and filtering the signal with the enhanced target voice and the signal with the weakened target voice according to the updated adaptive filter coefficient to get a signal with reduced noise; and weighing a voice presence probability by the credibility, and enhancing the signal with reduced noise according to the weighed voice presence probability.

Abstract: At least first and second microphones with different frequency responses form part of a speech recognition system. The microphones are coupled to a processor that is configured to recognize a spoken word based on the microphone signals. The processor classifies the spoken word, and weights the signals from the microphones based on the classification of the spoken word.