Pitch Perfect, a public speaking helper
- Open the mobile app, Pitch Perfect, and position the phone so it is secured in front of you, roughly at eye level.
- Record a video of yourself speaking, making sure to look at the camera, work on your body language, and smile!
- Finish recording and the video will be analyzed.
- See your personalized feedback on: a) Body Language - Eye contact tracking, posture and gestures, facial emotion recognition. b) Audio - Tone analysis, pauses, filler words
Features: 1) Eye Contact Tracking: We used CLAHE Equalization to identify the iris radii on the face. Then we used the Watson Vision Recognition API to find the offset distance between the center of the two irises and the center of the face bounding box. We recorded the time when the difference between these central points was greater than a threshold distance. We timed the durations of staying within the threshold (keeping consistent eye contact) and the durations of switching eye contact and presented these stats and recommendations for the user. 2) Facial Emotion Recognition: We used the Microsoft Azure API to scan emotion of frames of the input video and plot the varying expressions over the course of the speech. We monitored the amount of time that the user was smiling and presented this info as well. 3) Posture and Gestures: We trained a classifier with the IBM Watson custom model framework based on confident and diffident hand gestures and postures and classified the video frames accordingly. We tell the user what percentage of the time their posture is confident and when it appears nervous. 4) Filler Words: With the Rev API, we converted user speech to text and counted filler words, which we inform the user of. 5) Pauses and Tone analysis: We record the pace of speech and count the time between words at the end of punctuation to find the average pauses and advise the user on slowing down/speeding up. We used the IBM Watson Sentiment Analysis tool to monitor and inform the user of the tone of the speech.
Front-end Features: 1) Record Video 2) Annotate text with filler words and pauses 3) Present tone and body language stats
Log in or sign up for Devpost to join the conversation.