Abstract: One embodiment of the present application sets forth a computer-implemented method that includes receiving, from a first microphone, a first input acoustic signal, generating a first audio spectrum from at least the first input acoustic signal, wherein the first audio spectrum includes a set of time-frequency bins, and selecting a first time-frequency bin from the set based on a first local space-domain distance (LSDD) computed for the first time-frequency bin.

Abstract: An audio assembly includes a microphone assembly, a controller, and a speaker assembly. The microphone assembly detects audio signals. The detected audio signals originate from audio sources located within a local area. Each audio source is associated with a respective beamforming filter. The controller determines beamformed data using the beamforming filters associated with each audio source and a relative contribution of each of the audio sources using the beamformed data. The controller generates updated beamforming filters for each of the audio sources based in part on the relative acoustic contribution of the audio source, a current location of the audio source, and a transfer function associated with audio signals produced by the audio source. The controller generates updated beamformed data using the updated beamforming filters and performs an action (e.g., via the speaker assembly) based in part on the updated beamformed data.

Abstract: A system performs an optimization algorithm to optimize two or more acoustic sensors of a microphone array. The system obtains an array transfer function (ATF) for a plurality of combinations of the acoustic sensors of the microphone array. In a first embodiment, the algorithm optimizes an active set of acoustic sensors on the eyewear device. The plurality of combinations may be all possible combinations of subsets of the acoustic sensors that may be active. In a second embodiment, the algorithm optimizes a placement of two or more acoustic sensors on an eyewear device during manufacturing of the eyewear device. Each combination of acoustic sensors may represent a different arrangement of the acoustic sensors in the microphone array. In each embodiment, the system evaluates the obtained ATFs and, based on the evaluation, selects a combination of acoustic sensors for the microphone array.

Abstract: An eyewear device includes an audio system. In one embodiment, the audio system includes a microphone array that includes a plurality of acoustic sensors. Each acoustic sensor is configured to detect sounds within a local area surrounding the microphone array. For a plurality of the detected sounds, the audio system performs a direction of arrival (DoA) estimation. Based on parameters of the detected sound and/or the DoA estimation, the audio system may then generate or update one or more acoustic transfer functions unique to a user. The audio system may use the one or more acoustic transfer functions to generate audio content for the user.

Abstract: A system and method to perform an estimation of a location of an active speaker in real time includes designating a microphone of an array of microphones as a reference microphone. The method includes storing a relative transfer function (RTF) for each microphone of the array of microphones other than the reference microphone associated with each potential location among potential locations as a set of stored RTFs, and obtaining a voice sample of the active speaker and obtaining a speaker RTF for each microphone of the array of microphones other than the reference microphone. The method also includes performing an RTF projection of the speaker RTF for each microphone on the set of stored RTFs, and determining one of the potential locations as the location of the active speaker based on the performing the RTF projection.

Abstract: Technical solutions are described for acoustic echo cancellation. An example method includes computing, by a beamformer, a plurality of updated beamformer filter coefficients, the beamformer filter coefficients computed adaptively to determine a speech signal from a plurality of input audio signals. Further, the method includes computing, by a transfer function estimator, a relative transfer function based on the updated beamformer filter coefficients and a plurality of present acoustic echo canceller coefficients. Further, the method includes adjusting a plurality of acoustic echo canceller coefficients using the relative transfer function, and generating an output speech signal, by the acoustic echo canceller, by cancelling echo components from the speech signal using the adjusted filter coefficients. Further, the method includes sending, by the acoustic echo canceller, the output speech signal to a far end speech device.

Abstract: A system and method to perform an estimation of a location of an active speaker in real time includes designating a microphone of an array of microphones as a reference microphone. The method includes storing a relative transfer function (RTF) for each microphone of the array of microphones other than the reference microphone associated with each potential location among potential locations as a set of stored RTFs, and obtaining a voice sample of the active speaker and obtaining a speaker RTF for each microphone of the array of microphones other than the reference microphone. The method also includes performing an RTF projection of the speaker RTF for each microphone on the set of stored RTFs, and determining one of the potential locations as the location of the active speaker based on the performing the RTF projection.