Borboleta: A Mobile Telehealth System

An Open Source Project - Mobile Telehealth Services for all

Speech Recognition

A general goal of this research line is the use of automatic speech recognition in the human/machine interaction. Using speech in combination with conventional input/output methods can reduce user attention on the system interface and allow for more natural communication. Voice can also be used to improve data collection in environments where it is not easy use conventional input methods.

Our first task is the development of smart navigation schemas on mobile devices using voice commands and multimodal interfaces. A smart navigation schema will allow healthcare professionals more time during a visit to focus on patient needs. The mobile device will include an embedded small-vocabulary speech recognition system for navigation.

Our second task is data mining support. Historically, records are handwritten on paper. Collecting records and detailed histories electronically via the Borboleta mobile system allows for speech-to-text conversion and better data mining input. We will implement a large vocabulary speech recognition system using a vocabulary expanded with medical terminology.

Our challenges include the following.

  • Mobile devices have small processing capacity and thus a small vocabulary system is a better fit. To collect histories the device only requires storage capacity for the record, which will be processed on a higher capacity computer with a large vocabulary system.
  • Noisy environments will affect data collection but can be addressed with microphones, screens and signal processing technology
  • There has been relatively small formal data collection of Brazilian Portuguese, which is needed to train statistical models of acoustics and language. Our efforts may include new data collection as well as data cleaning in order to take advantage of existing recorded data.

Downloads

Powered by Drupal - Design by Artinet