"I am now eight months old, and with another child I will be able to expand my base for crying," says Ana Laguna Pradas. Your comment is not a joke. During the first pregnancy in 2016, this data expert asked how to understand the needs of his son when he came into the world. Why are you crying? Will you be asleep, hungry, cramping? He then began to create a system that discovered what lies behind the crying of newborns.
The 31-year-old Laguna Pradas, in a dialogue with the people, says that his research began as a party in which motherhood and his knowledge of data and artificial intelligence meet. While his experiment is still a hobby (his activity is scientific data in BBVA Data & Analytics), he is now asking other parents to take part in exchanging their child's screams to improve the accuracy of their machine translation.
– How did you decide to create this system?
– Surgery about three years ago when I was pregnant with my first child. Like most parents in the standby mode, among the many questions he asked me: "Will I understand when I cry?". This hypothesis, together with my knowledge and analysis of the data, led me to try to determine whether artificial intelligence [IA] Under these circumstances, it could help. I have already worked in machine translation, while crying children are another means of communicating a human being. In addition, if we have already tried to create whale-wave algorithms or identify birds, and if Jane Goodall understands the language of chimps, why not try to translate a newborn with an algorithm?
– Do you recognize the inspiration for your machine?
– I first searched for scientific articles (not too much) and some existing applications. I found only a Chinese application with very bad comments. In addition, because of the visibility – a branch of linguistics that studies accent and language intonation – the Chinese algorithm can not achieve the same degree of accuracy in Spanish children. I recently discovered a very good American application (Chatterbaby), which detected pain because they use an infant database during vaccination.
What is your algorithm?
-Algorithm is an artificial neural network that draws a model before the tagged input data, which is capable of recognizing the reason for crying when no new data is displayed, and thus provides us with a probability. Specifically, the algorithm receives sound signals converted into spectrograms as input data and returns the probability that the new crying has never been seen hunger, sleep, and so on. It should be noted that the study is based on infants from 0 to 6 months who are crying for the primary need. In the seven or eight months they already experience other feelings, such as fear that the system does not take into account.
Now you get more screams.
Yes SoGooData, the non-governmental organization I have set up and whose main objective is to contribute with social projects with science of data and artificial intelligence. For example, issues of a social nature such as pollution or health. So, I use this website (sogooddata.org) to add a form whereby parents can give more tears. We're currently collecting an average of seven audio tracks a day. Fortunately, the mothers are very involved. Sound recordings are confirmed before they are transmitted to the algorithm, although the mother's mark is usually reliable from 85 to 90% of the time. Mothers and fathers, along with their instincts and because they know their child well, greatly contribute to labeling in context. This is proven, because when you cover this need, the child ceases to cry.
– What are the plans for your system? Will we finally know what a child wants when he cries?
Although I am expanding the database, I test other techniques of machine learning and deep learning to reduce the bias of my algorithm. Now I only have information about my child, and therefore, we can not make a generalization in order to create "Shazam children's crying". They will write scientific articles on this subject in order to oppose some hypotheses, such as nationality or gender. When the database is large enough to create a robust algorithm, the existing algorithm will retouch and the application will be developed.
Laguna Pradas expects the entire application to reach the market at the end of the year. Explain that it will work similar to audio-based sensing tools. "The application will already be equipped with a model and it will be more likely to show the reason for crying when it receives a new sound," promises.