Microsoft Technology Licensing, LLC

Other

Vinod Prakash

Chao He

Sriram Srinivasan

Yan Lu

Xiulian Peng

XiangYu Kong

Hosam Adel Khalil

Qin Li

Sang Kun Choe

Soren Skak Jensen

Warren Michael Lam

Hong Wang Sodoma

Robert Andreas Aichner

Disclosed in some examples are methods, systems, and machine-readable mediums which determine jitter buffer delay by inputting jitter buffer and currently observed network status information to a machine learned model that is trained using a reinforcement learning (RL) method. The model maps these inputs to an action to compress, stretch, or hold the jitter buffer delay, which is used by a recipient computing device to optimize the jitter buffer delay. The model may be trained using a simulator that uses network traces of past real streaming sessions (e.g., communication sessions) of users. By training the model through reinforcement learning, the model learns to make better decisions through reinforcement in the form of reward signals that reflect the performance of each decision.

Reinforcement learning for jitter buffer control

Techniques are described for managing synchronized jitter buffers for streaming data (e.g., for real-time audio and/or video communications). A separate jitter buffer can be maintained for each codec. For example, as data is received in network packets, the data is added to the jitter buffer corresponding to the codec that is associated with the received data. When data needs to be read, the same amount of data is read from each of the jitter buffers. In other words, at each instance where data needs to be obtained (e.g., for decoding and playback), the same amount of data is obtained from each of the jitter buffers. In addition, the multiple jitter buffers use the same playout timestamp that is synchronized across the multiple of jitter buffers.

Synchronized jitter buffers to handle codec switches

Example apparatus and methods concern performing stereo acoustic echo cancellation using a correlation based filter adaptation control approach and without using stereo de-correlation. An embodiment includes a stereo adaptive filter that produces an echo removed microphone signal from received audio signals. The embodiment includes a mono adaptive filter that produces an echo removed microphone signal from the received audio signals. A correlation detector determines a level of correlation between the received signals and provides a signal to an adaptive filter controller. The adaptive filter controller controls how the stereo adaptive filter and the mono adaptive filter adapt audio echo cancellation as a function of the correlation between the received signals. A signal selector may select for output the signal from either the stereo adaptive filter or the mono adaptive filter based, for example, on the power level of the signals.

Correlation based filter adaptation

Architecture that mitigates echo in voice communications using echo detection and adaptive management of attenuation by a non-linear processor (NLP). Suppression values provided by the NLP are determined based on echo detection and retained on a case-by-case basis to automatically increase or decrease the attenuation as needed. Feedback is incorporated that where the controls for the NLP attenuation is given by the amount of echo that remains, and this in turn affects the amount of echo that remains.

Mitigation of echo in voice communication using echo detection and adaptive non-linear processor

Mitigation of nonlinear echoes by dynamic range compression may be provided. An audio output may be evaluated to determine whether nonlinear echoes are interfering with an audio signal. In order to reduce such nonlinear echoes, a dynamic range compression algorithm may be applied to the audio signal prior to output. The signal may be modified according to a compression wave and/or subjected to an attenuation factor.

Mitigating the impact of nonlinear echos by dynamic range compression

Presence of echo in a conversation is determined through a search performed in the sub-band domain using independent short length adaptive filters across a range of sub-band and maximum expected echo delay values. The adaptive filters attempt to predict and cancel the echo in the microphone signal based on the content in the speaker signal. If substantial cancellation is achieved in any of the sub-band filters, echo presence in the microphone signal is determined.

Echo presence determination in voice conversations

Detecting at least one of an echo detector and a noise detector based on analysis of audio streams transmitted to and received from each endpoint of a conference. When certain characteristics of the respective audio streams for a given endpoint are classified as significant against certain criteria, a determination is made as to whether that endpoint is a source of echo and/or noise. Subsequent actions are taken to alert users and/or prevent broadcast of impaired signals.

Conference signal anomaly detection

A voice communication end device provides echo reduction when operating in a full duplex mode using acoustic echo cancellation, which includes periodic audio queue adjustments to account for drift. The end device performs a quality assessment by calculating consistency statistics for the queue adjustments to determine whether acoustic echo cancellation would be ineffective, and if so falls back to a half duplex mode using voice switching.

Timestamp quality assessment for assuring acoustic echo canceller operability

In one embodiment, a two-way telecommunication device may perform acoustic echo cancellation on incoming signals. An audio decoding module may produce an audio render signal. An audio capture interface may receive an audio capture signal. A short length adaptive filter may determine a time delay between the audio render signal and the audio capture signal by adaptively predicting a sub-band of the audio capture signal using a corresponding sub-band of the audio render signal.

Growing community of inventors

Redmond, WA, United States of America

Vinod Prakash