Abstract: Systems, methods, and devices detect wake signals included in audio signals. Methods include receiving a dataset including raw audio data, the raw audio data comprising a plurality of audio samples and associated metadata, and generating, using one or more processing elements, an augmented dataset based on the raw audio data, the augmented dataset comprising a plurality of annotations identifying types of raw audio data. Methods further include generating, using the one or more processing elements, a feature dataset by extracting features from the augmented dataset based, at least in part, on the plurality of annotations, and generating, using the one or more processing elements, a wake signal detection model based, at least in part, on the feature dataset, the wake signal detection model being a machine learning model trained based on the feature dataset.

Abstract: Described are techniques to recognize spoken wake word (WW) or command for human-machine interface using a speech recognition system that does not require any WW or command-matching speech data for training. The system uses the text or grapheme representation of the WW or commands for training before deployment. The technique includes receiving by the system a text representation of a target phrase of a target language. It includes training an acoustic model based on a speech database to distinguish speech signals according to acoustic units of the target language. The training of the acoustic model is independent of the target phrase, It includes constructing a recognition model based on the text representation of the target phrase and the acoustic model to recognize the target phrase in speech; and processing speech from a speaker based on the acoustic model and the recognition model to detect a presence of the target phrase.

Abstract: A system includes memory storing instructions and a processing device coupled to the memory. The processing device executes the instructions to: receive a noisy audio signal from an audio receiver; pass the noisy audio signal through a deep neural network (DNN) model to generate a mask of a magnitude spectrogram of the noisy audio signal; retrieve a clean-only audio spectra and a noise-only spectra of one or more frequencies that exist within the noisy audio signal; generate a loss function as a combination of the clean-only audio spectra multiplied by the mask and the noise-only spectra multiplied by the mask; and train the DNN model while minimizing the loss function to generate a trained DNN model useable in audio noise suppression and dereverberation.

Abstract: Methods and systems for receiving a trained machine learning model, receiving a test dataset, wherein the test dataset is used to evaluate the trained machine learning model, generating, based on the test dataset and the trained machine learning model, one or more suggested modifications to at least one aspect of configuring and training of the trained machine learning model, and applying the one or more suggested modifications to at least one aspect of configuring and training of the trained machine learning model.

Abstract: A system includes memory storing instructions and a processing device coupled to the memory. The processing device executes the instructions to: receive a noisy audio signal from an audio receiver; pass the noisy audio signal through a deep neural network (DNN) model to generate a mask of a magnitude spectrogram of the noisy audio signal; retrieve a clean-only audio spectra and a noise-only spectra of one or more frequencies that exist within the noisy audio signal; generate a loss function as a combination of the clean-only audio spectra multiplied by the mask and the noise-only spectra multiplied by the mask; and train the DNN model while minimizing the loss function to generate a trained DNN model useable in audio noise suppression and dereverberation.

Abstract: Methods and systems for generating balanced data sets for speech-based discriminative tasks. The disclosed method includes, among other things, generating, based on a plurality of natural speech recordings, a synthetic speech data set, modifying, based on language science resources, the synthetic speech data set, and generating, based on the modified synthetic speech data set and the plurality of natural speech recordings, a balanced data set for training a discriminative model to perform a speech-based discriminative task.

Abstract: Methods and systems for a machine learning model architecture for speech enhancement system. The disclosed machine learning model architecture includes, among other things, an encoder, a decoder, and a bottleneck disposed between the encoder and the decoder. The encoder includes a plurality of encoder layers, and the decoder includes a plurality of decoder layers. Each encoder layer is connected to a corresponding decoder layer via a skip connection. Each encoder layer includes a Res2Net and a squeeze-and-excitation (SE) block. The bottleneck includes a first gated recurrent unit (GRU) layers and a second GRU layer.

Abstract: Systems, methods, and devices detect wake signals included in audio signals. Methods include receiving a dataset including raw audio data, the raw audio data comprising a plurality of audio samples and associated metadata, and generating, using one or more processing elements, an augmented dataset based on the raw audio data, the augmented dataset comprising a plurality of annotations identifying types of raw audio data. Methods further include generating, using the one or more processing elements, a feature dataset by extracting features from the augmented dataset based, at least in part, on the plurality of annotations, and generating, using the one or more processing elements, a wake signal detection model based, at least in part, on the feature dataset, the wake signal detection model being a machine learning model trained based on the feature dataset.

Abstract: Techniques for non-linear acoustic echo cancellation are described herein. In an embodiment, a system comprises a loudspeaker, a microphone array, a spatial filtering logic with a spatial filter, an acoustic echo canceller (AEC) logic and an adder logic block. The spatial filtering logic is configured to generate a spatially-filtered signal by applying the spatial filter using a reference signal sent to the loudspeaker and a multi-channel microphone signal from the microphone array. The generated spatially-filtered signal carries both linear echo and non-linear echo that are included in the multi-channel microphone signal. The AEC logic is configured to apply a linear adaptive filter using the spatially-filtered signal to generate a cancellation signal that estimates both the linear echo and the non-linear echo of the multi-channel microphone signal. The adder logic block is configured to generate an output signal based on the cancellation signal.

Abstract: Techniques for non-linear acoustic echo cancellation are described herein. In an embodiment, a system comprises a loudspeaker, a microphone array, a spatial filtering logic with a spatial filter, an acoustic echo canceler (AEC) logic and an adder logic block. The spatial filtering logic is configured to generate a spatially-filtered signal by applying the spatial filter using a reference signal sent to the loudspeaker and a multi-channel microphone signal from the microphone array. The generated spatially-filtered signal carries both linear echo and non-linear echo that are included in the multi-channel microphone signal. The AEC logic is configured to apply a linear adaptive filter using the spatially-filtered signal to generate a cancellation signal that estimates both the linear echo and the non-linear echo of the multi-channel microphone signal. The adder logic block is configured to generate an output signal based on the cancellation signal.

Abstract: Techniques for acoustic echo cancellation are described herein. In an example embodiment, a system comprises a speaker, a microphone array with multiple microphones, a beamformer (BF) logic and an acoustic echo canceller (AEC) logic. The speaker is configured to receive a reference signal. The BF logic is configured to receive audio signals from the multiple microphones and to generate a beamformed signal. The AEC logic is configured to receive the beamformed signal and the reference signal. The AEC logic is also configured to compute a vector of bias coefficients multiple times per time frame, to compute a background filter coefficient based on the vector of bias coefficients, to apply a background filter to the reference signal and the beamformed signal based on the background filter coefficient, to generate a background cancellation signal, and to generate an output signal based at least on the background cancellation signal.

Abstract: Techniques for acoustic echo cancellation are described herein. In an example embodiment, a system comprises a speaker, a microphone array with multiple microphones, a beamformer (BF) logic and an acoustic echo canceller (AEC) logic. The speaker is configured to receive a reference signal. The BF logic is configured to receive audio signals from the multiple microphones and to generate a beamformed signal. The AEC logic is configured to receive the beamformed signal and the reference signal. The AEC logic is also configured to compute a vector of bias coefficients multiple times per time frame, to compute a background filter coefficient based on the vector of bias coefficients, to apply a background filter to the reference signal and the beamformed signal based on the background filter coefficient, to generate a background cancellation signal, and to generate an output signal based at least on the background cancellation signal.