Embedded sound recognition as a trigger for smart appliances
Audio Analytic licenses an embedded software platform called Artificial Audio Intelligence (ai3), that recognizes and responds to sounds within the home. Users can choose which sound profiles to activate on the device, out of existing sound profiles provided by Audio Analytic.
For the last seven years, the company has been recording and analysing the audio patterns of hundreds of different sounds susceptible to be used as trigger events, such as alarm sounds, breaking glass, barking dogs, baby cries.
“We reduce a sound in its core constituents the size of a small text file that describes to our ai3 (Artificial Audio Intelligence) engine all the sound’s varying types so it can recognize it in any context”, Audio Analytic’s CEO Chris Mitchell told eeNews Europe.
The idea is to have sound profiles embedded in smart devices so these can send notifications to end-users or connect to other devices and trigger their operation (such as video recording, taking a picture, establishing a voice communication etc…)
“We’ve built up a large collection of sounds with which we can train our artificial algorithms. But another aspect of learning is that once a device is installed, it constantly adapts to its acoustic environment, analysing the noise environment to remove constant noise, so it remains highly accurate while only requiring a small footprint”, explained Mitchell.
In effect, by analysing the noise environment, the ai3 engine can recognize a normative sound pattern in a given place, which can then become the basis for creating an alert whenever anomalous sound events occur.
When ai3 recognizes a given sound, it connects to the device Event Management Service (EMS) to respond as required by the use case. The platform also offers an audio clipping service where a short clip of the sound that triggered an ai3 alert is sent to the absent user for validation and executive action. Data of ai3 alerts over time can be served to the end-user by the OEM or service provider as an additional tracking service.
The key difference with many other sound or voice recognition solutions today is that the ai3 engine and multiple sound profiles can be fully embedded on very low cost devices, without requiring a connection to the cloud for its operation or sound validation. Sounds are recognized locally in real time, without having to stream audio to the cloud (and all the privacy issues of listening into a home).
The startup charges royalties per device per sound in relation with shipping volumes. It has developed a small evaluation kit running on a Raspberry Pi device and a full software development kit which allows engineers to embed sound recognition straight into existing products.
Audio Analytic – www.audioanalytic.com
Related articles:
Researchers unlock microphones to inaudible audio
Cognitive hearing aid puts DNNs to work with the wearer’s brain
Audio development kit improves voice recognition rates for H2M applications
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
