Abstract: A method includes obtaining, at a first conference endpoint device, spoken command data representing a spoken command detected by the first conference endpoint device during a teleconference between the first conference endpoint device and a second conference endpoint device. The method further includes generating modified spoken command data by inserting a spoken phrase into the spoken command. The method further includes transmitting the modified spoken command data to a natural language service.

Abstract: A method includes, during a teleconference between a first audio input/output device and a second audio input/output device, receiving, at an analysis and response device, a signal indicating a spoken command, the spoken command associated with a command mode. The method further includes, in response to receiving the signal, generating, at the device, a reply message based on the spoken command, the reply message to be output to one or more devices selected based on the command mode. The one or more devices includes the first audio input/output device, the second audio input/output device, or a combination thereof.

Abstract: A method includes, during a teleconference between a first audio input/output device and a second audio input/output device, receiving, at an analysis and response device, a signal indicating a spoken command, the spoken command associated with a command mode. The method further includes, in response to receiving the signal, generating, at the device, a reply message based on the spoken command, the reply message to be output to one or more devices selected based on the command mode. The one or more devices includes the first audio input/output device, the second audio input/output device, or a combination thereof.

Abstract: A device includes one or more processor devices, an audio sensor, a speaker device, and one or more memory devices. The audio sensor is configured to generate an audio stream. The one or more memory devices store data including data identifying a wake phrase and instructions. The instructions are executable by the one or more processor devices to initiate, as a result of detecting the wake phrase in the audio stream and determining that the wake phrase is invalid, presentation via the speaker device of audio output identifying a valid wake input. The instructions are further executable by the one or more processors to initiate, as a result of detecting the valid wake input, communication of a portion of the audio stream to a natural language processing service.

Abstract: Traditional audio feedback elimination systems may attempt to reduce the effect of the audio feedback by simply scaling down the audio volume of the signal frequencies that are prone to howling. Other traditional feedback elimination systems may also employ adaptive notch filtering to detect and “notch” the so-called “singing” or “howling” frequencies as they occur in real-time. Such devices may typically have several knobs and buttons needing tuning, for example: the number of adaptive parametric equalizers (PEQs) versus fixed PEQs; attack and decay timers; and/or PEQ bandwidth. Rather than removing the singing frequencies with PEQs, the devices described herein attempt to holistically model the feedback audio and then remove the entire feedback signal. Two advantages of the devices described herein are: 1.) the system can operate at a much larger loop-gain (and hence with a much higher loudspeaker volume); and 2) setup is greatly simplified (i.e., no tuning knobs or buttons).

Abstract: A device includes one or more processor devices, an audio sensor, a speaker device, and one or more memory devices. The audio sensor is configured to generate an audio stream. The one or more memory devices store data including data identifying a wake phrase and instructions. The instructions are executable by the one or more processor devices to initiate, as a result of detecting the wake phrase in the audio stream and determining that the wake phrase is invalid, presentation via the speaker device of audio output identifying a valid wake input. The instructions are further executable by the one or more processors to initiate, as a result of detecting the valid wake input, communication of a portion of the audio stream to a natural language processing service.

Abstract: Traditional audio feedback elimination systems may attempt to reduce the effect of the audio feedback by simply scaling down the audio volume of the signal frequencies that are prone to howling. Other traditional feedback elimination systems may also employ adaptive notch filtering to detect and “notch” the so-called “singing” or “howling” frequencies as they occur in real-time. Such devices may typically have several knobs and buttons needing tuning, for example: the number of adaptive parametric equalizers (PEQs) versus fixed PEQs; attack and decay timers; and/or PEQ bandwidth. Rather than removing the singing frequencies with PEQs, the devices described herein attempt to holistically model the feedback audio and then remove the entire feedback signal. Two advantages of the devices described herein are: 1.) the system can operate at a much larger loop-gain (and hence with a much higher loudspeaker volume); and 2) setup is greatly simplified (i.e., no tuning knobs or buttons).

Abstract: A method includes obtaining, at a first conference endpoint device, spoken command data representing a spoken command detected by the first conference endpoint device during a teleconference between the first conference endpoint device and a second conference endpoint device. The method further includes generating modified spoken command data by inserting a spoken phrase into the spoken command. The method further includes transmitting the modified spoken command data to a natural language service.

Abstract: Traditional audio feedback elimination systems may attempt to reduce the effect of the audio feedback by simply scaling down the audio volume of the signal frequencies that are prone to howling. Other traditional feedback elimination systems may also employ adaptive notch filtering to detect and “notch” the so-called “singing” or “howling” frequencies as they occur in real-time. Such devices may typically have several knobs and buttons needing tuning, for example: the number of adaptive parametric equalizers (PEQs) versus fixed PEQs; attack and decay timers; and/or PEQ bandwidth. Rather than removing the singing frequencies with PEQs, the devices described herein attempt to holistically model the feedback audio and then remove the entire feedback signal. Two advantages of the devices described herein are: 1.) the system can operate at a much larger loop-gain (and hence with a much higher loudspeaker volume); and 2) setup is greatly simplified (i.e., no tuning knobs or buttons).

Abstract: A method includes, during a teleconference between a first audio input/output device and a second audio input/output device, receiving, at an analysis and response device, a signal indicating a spoken command, the spoken command associated with a command mode. The method further includes, in response to receiving the signal, generating, at the device, a reply message based on the spoken command, the reply message to be output to one or more devices selected based on the command mode. The one or more devices includes the first audio input/output device, the second audio input/output device, or a combination thereof.

Abstract: Systems and methods for holistically modelling audio feedback and removing the entire feedback signal corresponding thereto. The systems can operate at a much larger loop-gain (and hence with a much higher loudspeaker volume), than those conventional systems which seek to remove singing frequencies with PEQs. The systems are an improvement over traditional audio feedback elimination systems which attempt to reduce the effect of the audio feedback by simply scaling down the audio volume of the signal frequencies that are prone to howling, and those feedback elimination systems which simply employ adaptive notch filtering to detect and “notch” the so-called “singing” or “howling” frequencies as they occur in real-time. Such devices may typically have several knobs and buttons needing tuning, for example: the number of adaptive parametric equalizers (PEQs) versus fixed PEQs; attack and decay timers; and/or PEQ bandwidth.

Abstract: A multimedia conferencing system includes a loud speaker system, one or more microphones for receiving a local audio signal and a remote audio signal, a state machine and an echo canceller that operates in conjunction with two reference signals to remove substantially all of a feedback signal component in the local audio signal that results from reinforcing and playing the local audio signal over the loud speaker system. The state machine operates to detect that only the local audio is active, and if so controls the operation of the echo canceller such that only the feedback component of the local audio signal is removed and the local audio signal is not suppressed.

Abstract: A multiplexed microphone signal with multiple signal processing paths is disclosed. Each signal processing path has it own priority and other characteristics. A signal path is selected based on the application of the processed signal. Similar processes within different paths may be shared to reduce computation workload.

Abstract: A multiplexed microphone signal with multiple signal processing paths is disclosed. Each signal processing path has it own priority and other characteristics. A signal path is selected based on the application of the processed signal. Similar processes within different paths may be shared to reduce computation workload.

Abstract: A method and apparatus for detecting a singing frequency in a signal processing system using two neural-networks is disclosed. The first one (a hit neural network) monitors the maximum spectral peak FFT bin as it changes with time. The second one (change neural network) monitors the monotonic increasing behavior. The inputs to the neural-networks are the maximum spectral magnitude bin and its rate of change in time. The output is an indication whether howling is likely to occur and the corresponding singing frequency. Once the singing frequency is identified, it can be suppressed using any one of many available techniques such as notch filters. Several improvements of the base method or apparatus are also disclosed, where additional neural networks are used to detect more than one singing frequency.

Abstract: Methods and devices for improving the intelligibility of audio in a teleconferencing unit. Multiple microphones and multiple audio channels are used, in which only the best microphones are selected to represent each audio channel. Multiple microphones signals may be mixed according to microphones' positions in a room to form a single signal to represent one audio channel. The audio signal may be further processed to effectuate other features.

Abstract: A system and method for removing noise from a signal containing speech (or a related, information carrying signal) and noise. A speech or voice activity detector (VAD) is provided for detecting whether speech signals are present in individual time frames of an input signal. The VAD comprises a speech detector that receives as input the input signal and examines the input signal in order to generate a plurality of statistics that represent characteristics indicative of the presence or absence of speech in a time frame of the input signal, and generates an output based on the plurality of statistics representing a likelihood of speech presence in a current time frame; and a state machine coupled to the speech detector and having a plurality of states. The state machine receives as input the output of the speech detector and transitions between the plurality of states based on a state at a previous time frame and the output of the speech detector for the current time frame.

Abstract: An adaptive filter is provided featuring a speech spectrum estimator receiving as input an estimated spectral magnitude signal for a time frame of the input signal and generating an estimated speech spectral magnitude signal representing estimated spectral magnitude values for speech in a time frame. A spectral gain generator receives as input the estimated spectral magnitude signal and the estimated speech spectral magnitude signal and generates as output an initial spectral gain signal that yields an estimate of speech spectrum in a time frame of the input signal when the initial spectral gain signal is applied to the spectral signal. A spectral gain modifier receives as input the initial spectral gain signal and generates a modified gain signal by limiting a rate of change of the initial spectral gain signal with respect to the spectral gain over a number of previous time frames. The modified gain signal is then applied to the spectral signal, which is then converted to its time domain equivalent.