Abstract: In one embodiment, a method includes accessing a window of audio that includes audio signals. The method further includes determining, for each of the audio signals, a power of that signal relative to a reference audio signal; determining, for each of the audio signals and based on the determined relative power of that signal, a gain to apply to the audio signal relative to the reference audio signal, where the gain depends on frequency and on directionality relative to a listener; and encoding the audio so that bandwidth is allocated based on directionality relative to the listener by allocating an amount of bandwidth to each of the audio signals based on their respective frequency-dependent and directionality-dependent determined gains.

Abstract: One embodiment provides a method comprising optimizing one or more stimuli parameters by applying machine learning to training data. The method further comprises determining, based on the one or more optimized stimuli parameters, stimuli for simultaneously exciting a plurality of speakers within a spatial area. The stimuli has a shortest possible duration that is accurate for simultaneous deconvolution of a plurality of impulse responses of the plurality of speakers. The method further comprises simultaneously exciting the plurality of speakers by providing the stimuli to the plurality of speakers at the same time for reproduction. The method further comprises simultaneously deconvolving the plurality of impulse responses based on the stimuli and one or more measurements of sound recorded during the reproduction and arriving at one or more microphones within the spatial area.

Abstract: In example implementations, an apparatus is provided. The apparatus includes an adaptive filter and a double talk detector in communication with the adaptive filter. The adaptive filter is to calculate a transfer function with coefficients for a particular time that is applied to an output signal of a microphone to cancel echoes caused by a reference signal in the output signal of the microphone. The double talk detector is to determine a peak of the coefficients, detect double talk based on a location of the peak of the coefficients, and transmit a pause signal to the adaptive filter in response to detection of the double talk, wherein the pause signal is to pause a calculation of updates to the coefficients by the adaptive filter.

Abstract: One embodiment provides a computer-implemented method that includes creating, during content production, an audio object and metadata associated with the audio object based on a motion vector analysis of an object in one or more image frames in a video. The method can include, during the content production, inserting the audio object and the metadata associated with the audio object into at least one of an audio encoder or a video encoder. The method can include, during content playback, rendering the audio object, without image frame analysis, based on decoding the audio object and parsing the metadata associated with the audio object.

Abstract: One embodiment provides a computer-implemented method that includes sending a stimulus signal to a loudspeaker. A measurement signal is received via a microphone. The stimulus signal is transformed into a stimulus time-frequency representation. The measured signal is transformed into a measured time-frequency representation. At least one frequency value is selected between the stimulus time-frequency representation and the measured time-frequency representation. Correlation analysis is performed using the selected at least one frequency value. Based on the correlation analysis, a statistical mode is determined to produce a start-time of the stimulus signal.

Abstract: Embodiments are described for a method of rendering audio for playback through headphones comprising receiving digital audio content, receiving binaural rendering metadata generated by an authoring tool processing the received digital audio content, receiving playback metadata generated by a playback device, and combining the binaural rendering metadata and playback metadata to optimize playback of the digital audio content through the headphones.

Abstract: One embodiment provides a method of signal normalization. The method comprises receiving an input content with a corresponding audio signal, and extracting loudness metadata from an audio signal corresponding to the input content. The method further comprises estimating, using a machine learning model, a peak-level amplitude based on the loudness metadata. The peak-level amplitude represents a maximum linear amplitude of the audio signal over an entire duration of the input content. The method further comprises determining a gain based at least on the peak-level amplitude, and applying the gain to the audio signal. The resulting gain-scaled audio signal is provided to one or more speakers coupled to or integrated in an electronic device for audio playback.

Abstract: Embodiments are described for a method of rendering audio for playback through headphones comprising receiving digital audio content, receiving binaural rendering metadata generated by an authoring tool processing the received digital audio content, receiving playback metadata generated by a playback device, and combining the binaural rendering metadata and playback metadata to optimize playback of the digital audio content through the headphones.

Abstract: One embodiment provides a computer-implemented method that includes determining directional sounds from a content mix using a machine learning unmixing model. The directional sounds are panned in an upmixed signal. Signal-dependent upmixing gains for specific frequency bins are computed on a frame-basis using a machine learning model for the upmixed signal. Dedicated voice clarity gains are computed using a hearing impairment model for multiple hearing-impaired profiles for achieving dialog enhancement. The signal dependent upmixing gains and voice clarity gains are transmitted as metadata with a downmixed signal representing the content mix.

Abstract: One embodiment provides a computer-implemented method that includes creating, during content production, an audio object and metadata associated with the audio object based on a motion vector analysis of an object in one or more image frames in a video. The method can include, during the content production, inserting the audio object and the metadata associated with the audio object into at least one of an audio encoder or a video encoder. The method can include, during content playback, rendering the audio object, without image frame analysis, based on decoding the audio object and parsing the metadata associated with the audio object.

Abstract: One embodiment provides a method of audio upmixing comprising performing video scene analysis by segmenting visual objects from video frames of a video, and performing audio analysis by extracting audio signals from an audio corresponding to the video. The method further comprises determining whether any of the audio signals correspond to any of the visual objects, and estimating a video-based trajectory of a visual object if the visual object is in motion and transitions from on-screen to off-screen, or vice versa, during the video. The method further comprises positioning an audio trajectory of an audio signal from at least one speaker associated with the display to at least one other speaker associated with providing surround sound. The audio trajectory is automatically matched with the video. The audio signal is delivered to the at least one speaker and the at least one other speaker for audio reproduction during the presentation.

Abstract: One embodiment provides a computer-implemented method that includes utilizing text information obtained from a title of a media content item and a trainable model for improving accuracy for classification of the media content item. The trainable model is utilized using a sequence of text to numeric-vector embeddings for classification of the media content item. At least one of a word embedding model parameter or a latent semantic analysis dimension is jointly optimized using the text information, and a classifier model for maximizing accuracy of the classification of the media content item.

Abstract: One embodiment provides a computer-implemented method that includes implementing a customizable compressor for at least one sidechain processing associated with a loudspeaker. Machine learning is applied to automatically tune one or more parameters of the at least one sidechain processing. One or more channels are extracted, including a low-frequency effects (LFE) channel, for nonlinear signal synthesis. A proportional power-sum-based mix-in of an LFE sidechain channel is applied into a non-LFE sidechain. The LFE sidechain channel is maintained within a specified threshold of being level, before and after nonlinear signal synthesis.

Abstract: Acoustic echo cancellation for a video conference system is described. A location of a person in a room can be determined. An audio signal received from the location of the person can be captured using beamforming. An acoustic echo cancellation parameter can be determined based in part on the audio signal captured from the location of the person. Acoustic echo cancellation can be performed on the audio signal using the acoustic echo cancellation parameter.

Abstract: Embodiments are described for a method of rendering audio for playback through headphones comprising receiving digital audio content, receiving binaural rendering metadata generated by an authoring tool processing the received digital audio content, receiving playback metadata generated by a playback device, and combining the binaural rendering metadata and playback metadata to optimize playback of the digital audio content through the headphones.

Abstract: One embodiment provides a method comprising optimizing one or more stimuli parameters by applying machine learning to training data. The method further comprises determining, based on the one or more optimized stimuli parameters, stimuli for simultaneously exciting a plurality of speakers within a spatial area. The stimuli has a shortest possible duration that is accurate for simultaneous deconvolution of a plurality of impulse responses of the plurality of speakers. The method further comprises simultaneously exciting the plurality of speakers by providing the stimuli to the plurality of speakers at the same time for reproduction. The method further comprises simultaneously deconvolving the plurality of impulse responses based on the stimuli and one or more measurements of sound recorded during the reproduction and arriving at one or more microphones within the spatial area.

Abstract: One embodiment provides a method comprising determining stimuli for simultaneously exciting a plurality of speakers within a spatial area. The method further comprises simultaneously exciting the plurality of speakers by providing the stimuli to the plurality of speakers at the same time for reproduction. The method further comprises recording, during the reproduction, one or more measurements of sound arriving at one or more microphones within the spatial area. The method further comprises simultaneously deconvolving a plurality of impulse responses of the plurality of speakers based on the stimuli and the one or more measurements.

Abstract: Embodiments are described for a method of rendering audio for playback through headphones comprising receiving digital audio content, receiving binaural rendering metadata generated by an authoring tool processing the received digital audio content, receiving playback metadata generated by a playback device, and combining the binaural rendering metadata and playback metadata to optimize playback of the digital audio content through the headphones.

Abstract: In example implementations, an apparatus is provided. The apparatus includes an adaptive filter and a double talk detector in communication with the adaptive filter. The adaptive filter is to calculate a transfer function with coefficients for a particular time that is applied to an output signal of a microphone to cancel echoes caused by a reference signal in the output signal of the microphone. The double talk detector is to determine a peak of the coefficients, detect double talk based on a location of the peak of the coefficients, and transmit a pause signal to the adaptive filter in response to detection of the double talk, wherein the pause signal is to pause a calculation of updates to the coefficients by the adaptive filter.

Abstract: In one example in accordance with the present disclosure, a system is described. The system includes a processor to perform spatial audio processing on a received audio signal and an audio interface to connect an audio output device to a computing device. The system also includes a controller. The controller determines a spatial audio processing capability of the audio output device and disables spatial audio processing on one of the audio output device and the processor based on a determination of the spatial audio processing capability of the audio output device.