Catch-A-Cry
Informatics Capstone Poster/ 2021
Doctors and parents in clinical trials face significant difficulties in tracking babies’ cries. The current method, handwritten crying logs, are inaccurate and offer little insight for clinical personnel. This project aims to use machine learning algorithms to more accurately identify and track cries. Hours of crying data was gathered and transformed into t-Distributed Stochastic Neighbor Embedding, Support Vector Machine, and k-Nearest Neighbor algorithms, visually grouping audio clips by sound similarity. An audio debugger determines the accuracy of the groupings. With continued training and testing, the algorithms will be able to engage binary classification of cries.
CAC_Poster | |
File Size: | 480 kb |
File Type: |
Key Features
Audio Conversion + Display
With audio data that may be provided, we able to convert them into specified formats (kHz, etc.) and run it through an algorithm (tSNE) that helps plot the audio files at a high-dimension visualization. This is able to produce similar data points into a cluster via a classifier (SVM) in which we can specify what certain audios are classified as: hunger, discomfort, pain, etc. In addition to this, we are able to produce a similar graph to pre-labeled audio data that may have been used in clinical trials by a KNN algorithm in which we can train data and also help specify frequency of audio points to help for classification above. Data Playback
After we have run the data through conversion, we're able to listen to the sound files through an interactive graph (tSNE) displayed on a web application. Spectrogram Visualization
From the data points we have then we can later on play the audio, and help showcase the characteristics in the cry, based on tone, pitch visually through the spectrogram. Below: |
Presentation Video
ObvioHealthObvioHealth is a Virtual Research Organization (VRO) leading the long overdue digitization of health innovation. The organization launched in 2017, leveraging technological innovation to tackle the inefficiencies in the trial process. However, in order to achieve success, the organization has brought a fusion of experienced clinical researchers and pioneering health technologists for seamless clinical trials.
The mission that ObvioHealth holds is to lead the virtual clinical trial revolution-- pioneering technologies and re-engineering processes to make health research easier, safer and more accurate. With this goal in mind, the Director of Research (DOR), Brian Cohn, Ph.D. has been leading the Catch-A-Cry Team, in order to help define crying as a clinical endpoint - through machine learning algorithm that can listen and identify infant crying. |
Catch-A-Cry Team
Project Status
The Catch-A-Cry Project by ObvioHealth is a University of Washington Information School Capstone Project for the Winter/Spring 2021 quarters. This project is a requirement for graduation from the Informatics major, at the University of Washington, Seattle. After the capstone completion, this project will be handed off to ObvioHealth for further development and testing within its organization.
For more information regarding ObvioHealth and Catch-A-Cry, please contact: Brian Cohn Ph.D. at [email protected]
For more information regarding ObvioHealth and Catch-A-Cry, please contact: Brian Cohn Ph.D. at [email protected]