Artificial intelligence is not just science fiction á la i-Robot or Terminator, but we integrate it into our everyday lives as a matter of course: Siri, Alexa, navigation apps, image recognition or language translation - all of these are based on artificial intelligence. At Livy, we also work with artificial intelligence and use AI to help our Smart Living Station recognize calls for help.
What is Artificial Intelligence?
Artificial intelligence (AI) is the ability of a system to mimic human abilities such as reasoning, learning, planning, and creativity. AI enables technical systems to perceive their environment, process data and solve problems to achieve a specific goal.
Using Alexa as an example, we know that asking the right questions leads to "real" conversations with the voice control and we get the feeling that Alexa understands us and has an answer for everything. Alexa voice assistance is a self-learning software that draws conclusions from all voice inputs (including misunderstood ones) with the aim of constantly improving the user's understanding.
Strong and weak AI
Artificial intelligence is divided into strong and weak AI.
Strong AI learns independently without human intervention and acquires its own creative and meaningful processes and knowledge in order to make decisions based on them and is thus similar to the human brain or can even be superior to it. So far, we have only seen strong AI in movies and it remains science fiction for the time being.
So far, there is only so-called weak or methodical AI in the real world, which is trained to recognize patterns in data and to derive decision rules from them (machine learning), so that they can subsequently make independent decisions.
Our Smart Living Station - which we affectionately call simply Livy - is also to become more intelligent with the help of AI and in the future will independently recognize calls for help and in appropriate situations immediately trigger an alarm or inform trusted persons without triggering false alarms in normal conversation situations. Of course, all of this will be done in compliance with data protection laws.
A fall in the bathroom or from the stairs - an accident at home can happen quickly and, especially for seniors and people in need of care, immediate help is important in such moments.
The Smart Living Station is designed to make everyday life even safer for people and to create a quick connection to their confidants in an alarm situation.
Thanks to the call for help recognition, users will be able to summon help without having to manually trigger a classic home emergency call system, for example. In the event of an alarm, community members are informed immediately, potentially saving the lives of thousands of people.
How do you train an artificial intelligence?
Anyone who believes that a smart function based on artificial intelligence can be implemented in just a few weeks is mistaken. Instead, it first requires a lot of research, the right parameters and a suitable training model with which to train the AI.
Just as we individually choose the most suitable method for learning for ourselves, machine learning also requires different methods and training models that we can apply to the different requirements of the AI.
Accordingly, Livy also needs a method to learn how to recognize a call for help. For this, we use the "supervised method" - supervised machine learning.
In order to use supervised learning meaningfully, data must be available with its associated result for a clearly defined goal or prediction. That is, the AI training model must be fed data, each of which is marked as correct or incorrect - depending on the AI's individual training goal. The system then learns via training through vast amounts of data to differentiate independently and take expected action accordingly.
Since few weeks, Livy has therefore been learning to distinguish acoustic positives from negatives using a training algorithm via a sound classification model. That is, to differentiate a real call for help from other words and sounds.
The goal for Livy is to identify a call for help and then notify community members so they can check on it and provide assistance if necessary. At the same time, Livy should not respond to normal conversations that mention help, nor to words that sound similar, so as not to trigger false alarms.
Supervised Learning: Livy learns the call for help
In addition to the method, a suitable training model is needed.
As mentioned above, in the case of Livy, this is a sound classification model that is specifically trained to identify certain sounds/words - in Livy, the call for help.
For this we have chosen an open source - i.e. a freely available and customizable model. There are quite a few of these on the Internet, and using them as a guide is a sensible, resource-saving and cost-efficient solution.
For the training itself, it needs acoustic positive and negative examples as a training data set, from which Livy learns.
This means: a large number of voice samples for the word help, but also for other (similar) words and sounds, from which Livy learns to distinguish, as the station would otherwise trigger a number of false alarms.
So the next step was to gather data:
As many different voices as possible from different people who speak the word help and many others.
Then our in-house AI specialist marked this data as correct or incorrect and entered it into the sound classification model as a training data set.
Currently, then, Livy is learning to hear and recognize the word help.
However, for the most efficient and reliable help call recognition, we need much more data. Therefore you can support Livy
Support the call for help recognition:
Leave us your comment and confirm your wish to participate. We will then contact you with an individual link that you can use to participate.
Using the link, you can easily enter various words directly from your smartphone, tablet or laptop without any special tools.
NOTE: All data is processed anonymously and forms the basis of a reliable call for help detection with Livy Alive for us. Every single recording helps us to further develop the function. It is not necessary to complete all levels. Thank you for your participation!
Participate and comment now.