Exotel aspires to be the one-stop solution for voice-based communication for businesses. Businesses typically have thousands of customer interactions every day. In India, most of this happens on the phone.
An interesting challenge that Exotel faced was offering businesses useful information and intelligence based on their conversations. They wanted to see how some of the smartest engineering minds would approach this problem.
Emotion detection from audio is an unsolved problem. Create a system that allows you to detect emotions from audio and flags conversations based on sentiments, such as, happiness, sadness, anger etc.
Exotel was started in 2011 with 3 people. Half a decade later, it is Southeast Asia’s largest company with more than 100 employees. It offers a cloud-based telephony platform that helps businesses communicate with its customers efficiently over calls and via SMS. With over 1300 customers, Exotel powers more than 3 million customer conversations everyday and has processed 1.2 billion calls in the past 5 years.
The participants were given sample audio files (training data set) and asked to use speech-recognition algorithms and machine learning to write code that would use the voice as an input and recognize the emotion behind it. The code would then be run across thousands of voice samples to choose the best model.