The problem SpeakInOut solves

This tool is aimed towards helping people overcome their fear of public speaking and face real life scenarios like interviews with heightened confidence. This year during our university placement, a lot of people were technically sound but lacked the necessary communication skills. Helping them improve on this aspect will open up new doors to a bright career. This app can be used by anyone, absolutely anyone! The tool records the video and audio of a person while speaking and performs effective analysis to help people improve their skills.
Now, while speaking in front of a big crowd people find it difficult to maintain eye contact. Eye contact is an important sign of confidence. The video analyses the eye gaze movement and detects the frequency at which a person looks away from the crowd. This gaze movement helps in calculating the confidence of a person. The audio is analysed to detect the number of filler words. Everyone has some filler words like "umm", "like", "actually" which they use often when they are short of words. Eliminating these words can help in improving the speech quality and help in capturing the undivided attention of audience.
A lot of times, people fear the follow up questions from the audience after their speech. This tool automatically generates some questions from the speech, so the user can get a feel of how tackling these questions would actually be like !
Sometimes people need to present a speech from a determined manuscript. Although this seems like a simpler task, people find it difficult to replicate it or end up looking too much at the manuscripts while speaking. This tool helps in measuring the similarity of speech to manuscript so the user can improve. The tool also provides a simulation of distraction which is a usually a norm in real life. The tool keeps a track of all the speeches and their analysis so users can sense their growth.

Challenges we ran into

• After several design iterations, we offloaded the heavier tasks for analysis of the quality of speech to background threads. Analysis of the speech recording was done asynchronously for capturing the relevant features needed to give insights to the user.
• We also faced some difficulties finding a good threshold value for defining eye contact with the crowd.- We faced several challenges in coming up with a robust solution to keep a track of recordings to capitalize the analytics from them. After several design solved the problem by using sqlite to store the data.