Power state control of a device performing detection of voice in audio data

A computing device, comprising:

• a wireless transmitter to exchange data with at least a wearable device;
• one or more processors coupled to the wireless transmitter;
• and
• a memory storing program instructions that, when executed by the one or more processors, cause the one or more processors to at least:
• receive, from the wearable device and via the wireless transmitter, user activity data indicating interaction between a user and the wearable device;
• receive collected audio data from the wearable device;
• select a monitoring schedule of a plurality of monitoring schedules for execution by the wearable device based at least in part on the user activity data and the collected audio data, the monitoring schedule to cause the wearable device to alternate between (i) a low power state where at least one of a collection or an analysis of audio data is disabled to reduce power consumption by the wearable device and (ii) an active power state where the collection and the analysis of the audio data is enabled, wherein the analysis of the audio data includes detection of a presence of a voice in the audio data;
• determine, from the wearable device, a delay time that extends from an initial time represented in the collected audio data to a detection time where the voice is first detected in the collected audio data;
• add at least part of the delay time to a length of time of a first duration included in the monitoring schedule to cause the wearable device to extend operation in the low power state during a subsequent duration to conserve available power of a battery of the wearable device;
• and
• transmit the monitoring schedule to the wearable device.

A computer-implemented method to be performed by a wearable device, the computer-implemented method comprising:

• setting a voice activity detector to operate in a low power state for a first duration in accordance with a monitoring schedule, the voice activity detector and monitoring schedule stored in memory of the wearable device, wherein the voice activity detector monitors for a voice included in audio data collected in part using a microphone in accordance with the monitoring schedule;
• setting the voice activity detector to operate in an active power state for a second duration in accordance with the monitoring schedule, the active power state to increase power consumption by the wearable device in comparison to operation in the low power state;
• collecting current audio data for at least the second duration;
• determine, by the voice activity detector, presence of the voice in the current audio data;
• determining, by the voice activity detector, a delay time that extends from an initial time represented in the current audio data to a detection time where the voice is first detected in the current audio data;
• and
• adding at least part of the delay time to a length of time of the first duration to cause the voice activity detector to extend operation of the voice activity detector in the low power state during a subsequent duration to conserve available power of a battery of the wearable device.