Recognizing hazard events based on sounds of in-home activities

1. A method comprising:

receiving at least an indication of audio data generated based on sound captured of activity of a user;

generating a sequence of audio event labels by processing the received audio data with a first machine learning model,

wherein the first machine learning model is configured to recognize and characterize predefined audio patterns using multiple audio recognition patterns, and

wherein each audio event label is generated by matching an audio recognition pattern in the received audio data with an audio recognition pattern of the multiple audio recognition patterns;

processing the sequence of audio event labels with a second machine learning model,

wherein the second machine learning model is trained to learn a baseline of audio event labels for activity and inactivity specific to the user;

detecting, using an output of the second machine learning model, an anomalous occurrence of an event based on a deviation between the sequence of audio event labels and the baseline of audio event labels; and

in response to detecting the anomalous occurrence of the event with the second machine learning model, generating an alert indicating occurrence of the event.

2. The method of claim 1 , wherein the sequence of audio event labels is for audio data in respective sequential and non-overlapping time windows.

3. The method of claim 1 , wherein the sequence of audio event labels is for audio data in overlapping time windows.

4. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting occurrence of the event at an unexpected time of day.

5. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting the event based on outputs of multiple monitoring devices configured to detect events in a time window in which the sound of the activity of the user occurred.

6. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting the event based on a temperature output by a temperature sensor of a monitoring device that captured the sound of the activity of the user.

7. The method of claim 1 :

wherein the second machine learning model is configured to identify a particular event type based on a location of a monitoring device that captures sounds of activity of the user, and

wherein the location includes a living room, a bathroom, or a bedroom of a house being monitored by the monitoring device.

8. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

failing to detect the event at an expected time.

9. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting that the event occurred at an unexpected time.

10. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting a non-verbal activity of the user.

11. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting a verbal activity of the user.

12. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting an anomalous pattern of the sequence of audio event labels; and

matching the anomalous pattern to an event type of multiple event types.

13. The method of claim 1 , wherein detecting the anomalous occurrence of the event comprises:

detecting a pattern in the sequence of audio event labels; and

comparing the detected pattern of the sequence of audio event labels to a baseline pattern of audio event labels,

wherein the detected pattern spans a first timeframe and the baseline pattern spans a second timeframe equal to or greater than the first timeframe.

14. A system, comprising:

one or more processors; and

one or more memories including instructions that, when executed by the processors,

cause the system to:

obtain audio data based on sounds captured of activity of a user;

process the audio data with a classification machine learning model configured to recognize and classify audio patterns to generate a sequence of audio event labels;

train an event detection machine learning model based on the sequence of audio event labels relative to predefined events,

wherein the event detection machine learning model is trained to identify occurrence of a plurality of events based on audio data;

process the sequence of audio event labels with the event detection machine learning model trained to learn a baseline of audio event labels for activity and inactivity specific to the user; and

detect, based on an output of the event detection machine learning model, an anomalous occurrence of an event based on a deviation between the sequence of audio event labels and the baseline of audio event labels.

15. The system of claim 14 , further comprising:

a transceiver configured to receive a representation of the sequence of audio event labels over a network from a device that captured the sounds of the activity of the user.

16. The system of claim 15 , wherein the representation is a graphical representation of sound captured over a predefined time period.

17. The system of claim 15 , wherein the audio data is a stream of audio data generated in real-time or near real-time based on the sounds captured by a microphone of a monitoring device.

18. A monitoring device comprising:

a microphone;

a network interface;

a processor; and

a memory including instructions that, when executed by the processor, cause the monitoring device to:

generate audio data based on sounds captured by the microphone, the sounds being of an activity of a user;

process the audio data with a first neural network to generate a sequence of audio event labels in temporal order,

wherein the first neural network is trained to detect segments that match predefined audio patterns;

process the sequence of audio event labels with a second neural network trained to learn a baseline of audio event labels for activity and inactivity specific to the user;

detect, based on an output of the second neural network, an anomalous occurrence of an event based on a deviation between the sequence of audio event labels and the baseline of audio event labels; and

in response to detecting the event, cause the network interface to communicate at least an indication of the event to an emergency service.

19. The monitoring device of claim 18 , wherein detecting the anomalous occurrence of the event comprises causing the monitoring device to:

fail to detect the event at an expected time.

20. The monitoring device of claim 18 , wherein detecting the anomalous occurrence of the event comprises causing the monitoring device to:

detect that the event occurred at an unexpected time.