Abstract: Low latency neural network models are provided that can be used for speech processing. The neural networks allow for real-time inference of CNN models without an increase in computer complexity or memory footprint. Buffers are used for upsampling, and the depth of the convolutions varies by frame number. In some examples, a condition is applied within the convolution block to determine a depth of convolutions based on the frame number. In some examples, the network includes multiple convolution sub-model blocks, each having a different depth, and a table is used to select the convolution sub-model block for each frame based on the frame number. The neural networks can be used for speech enhancement tasks such as dynamic noise suppression (DNS), blind source separation (BSS), and Self-Noise Silencers (SNS).

Abstract: Methods, apparatus, systems, and articles of manufacture to detect the location of sound sources external to computing devices are disclosed. An apparatus, to determine a direction of a source of a sound relative to a computing device, includes a cross-correlation analyzer to generate a vector of values corresponding to a cross-correlation of first and second audio signals corresponding to the sound. The first audio signal is received from a first microphone of the computing device. The second audio signal is received from a second microphone of the computing device. The apparatus also includes a location analyzer to use a machine learning model and a set of the values of the vector to determine the direction of the source of the sound.

Abstract: Techniques are provided for audio-based detection and tracking of an acoustic source. A methodology implementing the techniques according to an embodiment includes generating acoustic signal spectra from signals provided by a microphone array, and performing beamforming on the acoustic signal spectra to generate beam signal spectra, using time-frequency masks to reduce noise. The method also includes detecting, by a deep neural network (DNN) classifier, an acoustic event, associated with the acoustic source, in the beam signal spectra. The DNN is trained on acoustic features associated with the acoustic event. The method further includes performing pattern extraction, in response to the detection, to identify time-frequency bins of the acoustic signal spectra that are associated with the acoustic event, and estimating a motion direction of the source relative to the array of microphones based on Doppler frequency shift of the acoustic event calculated from the time-frequency bins of the extracted pattern.

Abstract: A method and system of audio processing encodes cochlear-simulating spike data into spectrogram data.

Abstract: A method and system of neural network dynamic noise suppression is provided for audio processing.

Abstract: Techniques are provided for dynamic noise suppression. A methodology implementing the techniques according to an embodiment includes generating a magnitude spectrum and a phase spectrum of an input audio signal comprising speech and dynamic noise. The method also includes employing a temporal convolution network (TCN) to generate a separation mask based on the magnitude spectrum. The TCN comprises depth-wise (DW) convolution layers, each DW convolution layer including a state buffer to store a number of previous states of the associated DW convolution layer. The number of stored previous states is based on a dilation factor of the associated DW convolution layer. The method further includes multiplying the separation mask with the magnitude spectrum to separate the speech from the dynamic noise to obtain a denoised magnitude spectrum. The method further includes reconstructing the input audio signal with reduced dynamic noise based on the denoised magnitude spectrum and the phase spectrum.

Abstract: Apparatus, systems, methods, and articles of manufacture are disclosed for acoustic signal processing adaptive to microphone distances. An example system includes a microphone to convert an acoustic signal to an electrical signal and one or more processors to: estimate a distance between a source of the acoustic signal and the microphone; select a signal processing mode based on the distance; and process the electrical signal in accordance with the selected processing mode.

Abstract: A system, method and computer readable medium for dynamic noise reduction in a voice call. The system includes an encoder having a short-time Fourier transform module to determine a magnitude spectrum and a phase spectrum of an input audio signal. The input audio signal includes speech and dynamic noise. A separator is coupled to the encoder. The separator comprises a temporal convolution network (TCN) used to develop a separation mask using the magnitude spectrum as input. The TCN is trained using a frequency SNR function used to calculate loss during training. A mixer is coupled to the separator to multiply the separation mask with the magnitude spectrum to separate the speech from the dynamic noise to obtain a denoise magnitude spectrum. The system also includes a decoder coupled to the mixer and the encoder. The decoder includes an inverse short-time Fourier transform module to reconstruct the input audio signal without the dynamic noise using the denoise magnitude spectrum and the phase spectrum.

Abstract: Methods, apparatus, systems, and articles of manufacture to detect the location of sound sources external to computing devices are disclosed. An apparatus, to determine a direction of a source of a sound relative to a computing device, includes a cross-correlation analyzer to generate a vector of values corresponding to a cross-correlation of first and second audio signals corresponding to the sound. The first audio signal is received from a first microphone of the computing device. The second audio signal is received from a second microphone of the computing device. The apparatus also includes a location analyzer to use a machine learning model and a set of the values of the vector to determine the direction of the source of the sound.

Abstract: Techniques are provided for audio-based detection and tracking of an acoustic source. A methodology implementing the techniques according to an embodiment includes generating acoustic signal spectra from signals provided by a microphone array, and performing beamforming on the acoustic signal spectra to generate beam signal spectra, using time-frequency masks to reduce noise. The method also includes detecting, by a deep neural network (DNN) classifier, an acoustic event, associated with the acoustic source, in the beam signal spectra. The DNN is trained on acoustic features associated with the acoustic event. The method further includes performing pattern extraction, in response to the detection, to identify time-frequency bins of the acoustic signal spectra that are associated with the acoustic event, and estimating a motion direction of the source relative to the array of microphones based on Doppler frequency shift of the acoustic event calculated from the time-frequency bins of the extracted pattern.

Abstract: A mechanism is described for facilitating automatic gain adjustment in audio systems according to one embodiment. A method of embodiments, as described herein, includes determining status of one or more of gain settings, mute settings, and boost settings associated with one or more microphones based on a configuration of a computing device including a voice-enabled device. The method may further comprise recommending adjustment of microphone gain based on the configuration and the status of one or more of the gain, mute, and boost settings, and applying the recommended adjustment of the microphone gain.

Abstract: An example apparatus for reducing noise in audio includes a preprocessor to receive audio input from a microphone and preprocess the audio input to generate preprocessed audio. The apparatus also includes an acoustic event detector to detect an acoustic event corresponding to a disturbance in the preprocessed audio. The apparatus further includes a noise reduction model selector to select a specific disturbance model based the detected acoustic event. The apparatus further includes a noise suppressor to attenuate components related to the disturbance in the preprocessed audio using the selected specific disturbance model to generate an enhanced audio with suppressed noise.

Abstract: System and techniques for automatic gain control for speech recognition are described herein. An audio signal may be obtained. A signal-to-noise ratio (SNR) may be derived from the audio signal. The SNR may be compared to a threshold. A stored gain value may be updated when the SNR is beyond the threshold and the stored gain value may be applied to a descendant (e.g., later) of the audio signal otherwise.

Abstract: Techniques are provided for pre-processing enhancement of a speech signal. A methodology implementing the techniques according to an embodiment includes performing de-reverberation processing on signals received from an array of microphones, the signals comprising speech and noise. The method also includes generating time-frequency masks (TFMs) for each of the signals. The TFMs indicate the probability that a time-frequency component of the signal associated with that TFM element includes speech. The TFM generation is based on application of a recurrent neural network to the signals. The method further includes generating steering vectors based on speech covariance matrices and noise covariance matrices. The TFMs are employed to filter speech components of the signals, for calculation of the speech covariance, and noise components of the signals for calculation of the noise covariance.

Abstract: A mechanism is described for facilitating multi-device reverberation estimation according to one embodiment. An apparatus of embodiments, as described herein, includes detection and capture logic to facilitate a microphone of a first voice-enabled device of multiple voice-enabled devices to detect a command from a user. The apparatus further includes calculation logic to facilitate a second voice-enabled device and a third voice-enabled device to calculate speech to reverberation modulation energy ratio (SRMR) values based on the command, where the calculation logic us further to estimate reverberation times (RTs) based on the SRMR values. The apparatus further includes decision and application logic to perform dereverberation based on the estimated RTs of the reverberations.

Abstract: A mechanism is described for facilitating multi-modal dereverberation in far-field audio systems according to one embodiment. A method of embodiments, as described herein, includes performing geometry estimation of a geographical space based on visuals of the space received from one or more cameras of a computing device. The method may further include computing reverberation time based on the geometry estimation that is further based on the visuals, and computing and applying dereverberation based on the reverberation time.

Abstract: A mechanism is described for facilitating simultaneous recognition and processing of multiple speeches from multiple users according to one embodiment. A method of embodiments, as described herein, includes facilitating a first microphone to detect a first speech from a first speaker, and a second microphone to detect a second speech from a second speaker. The method may further include facilitating a first beam-former to receive and process the first speech, and a second beam-former to receive and process the second speech, where the first and second speeches are at least received or processed simultaneously. The method may further include communicating a first output associated with the first speech and a second output associated with the second speech to the first speaker and the second speaker, respectively, using at least one of one or more speaker devices and one or more display devices.

Abstract: A mechanism is described for facilitating simultaneous recognition and processing of multiple speeches from multiple users according to one embodiment. A method of embodiments, as described herein, includes facilitating a first microphone to detect a first speech from a first speaker, and a second microphone to detect a second speech from a second speaker. The method may further include facilitating a first beam-former to receive and process the first speech, and a second beam-former to receive and process the second speech, where the first and second speeches are at least received or processed simultaneously. The method may further include communicating a first output associated with the first speech and a second output associated with the second speech to the first speaker and the second speaker, respectively, using at least one of one or more speaker devices and one or more display devices.

Abstract: Systems, apparatuses and methods for secure audio acquisition. The method includes receiving audio data via a digital microphone. The digital microphone outputs a single bit at a high sampling rate. The digital microphone output is converted to a full range audio signal. The full range audio signal is filtered to provide a band limited audio output that avoids capture of enough of a spectral range of speech for the speech to be intelligible.

Abstract: Techniques are provided for pre-processing enhancement of a speech signal. A methodology implementing the techniques according to an embodiment includes performing de-reverberation processing on signals received from an array of microphones, the signals comprising speech and noise. The method also includes generating time-frequency masks (TFMs) for each of the signals. The TFMs indicate the probability that a time-frequency component of the signal associated with that TFM element includes speech. The TFM generation is based on application of a recurrent neural network to the signals. The method further includes generating steering vectors based on speech covariance matrices and noise covariance matrices. The TFMs are employed to filter speech components of the signals, for calculation of the speech covariance, and noise components of the signals for calculation of the noise covariance.