Empowering Communities with Voice-enabled Technologies for Crisis Management

We asked Sarvodaya Emergency Coordinators how they communicate Almost none use the WWW (< 2%) Less than 50% of them use SMS Voice is the unanimous choice LIRNEasia conducted six country study on the use of mobile phones by the poor Empowering Communities with Voice-enabled Technologies for Crisis Management BY Nuwan Waidyanatha Senior Research Fellow A Mobile World Many communities are still lacking access to the Internet, they fall short in computer literacy and are underserved by non-latin scripting technologies! Thanks to the growing access to affordable mobile phones and interactive voice response technologies they can be part of the digital habitat. “We usually interconnect with other rescue and relief agencies through phones …” “We request for assistance from the head office through the phone ...” “It is easy to call … average citizens in the village don't use SMS ...” Sarvodaya Compassion in Action relief, rehabilitation, reconstruction over 1 million IDPs 2009 post war Sri Lanka Sarvodaya Shramadana Movement is Sri Lankas largest community-based grassroots development organization and they respond to all national and local disasters Sarvodaya Truck loads of Relief Rescue and Relief operations over 1 million affected 2011 Floods Sri Lanka Daily and Weekly reports are exchanged between the field coordinators and the head office on immediate resource needs Those reports are exchanged through Telephone calls, a few faxes and some are hand delivered Ideal Technology: Interactive Voice Response (IVR) integrated with Disaster Information Management System www.freedomfone.org Freedom Fone is a two-way IVR The user simply dials a phone number and navigates the voice menus to "leave-a-message" or listen to "content" in the local dialect Sahana is a web based localizable software The numerous modules are designed to manage humanitarian activities Information is captured and presented in categorical, visual, time-series and geo-spatial forms Both are Free and Open Source Software (FOSS) supported by global communities: www.freedomfone.org www.sahanafoundation.org Working with radio Now Imagine a crisis For exemple: radio staion can host the Sahana incident reportinga and missing persons registries for callers to voice their cries; those audio clips can be broadcasted for situational updates or response requests. Research Design and Evaluation Colombo (urban) Ratnapura (rural) Matara (urban) Principal: Lanka Jathika Sarvodaya Shramadana Sangamaya Sri Lanka's largest community development organization that responds to humanitarian services (www.sarvodaya.org) Hazard Information Hub @ Sarvodaya's Community Disaster Management Center, Moratuwa involved: three EM Coordinators and three Operators Four Districts: Colombo, Matara, Nuwara-eliya, Ratnapura, ~ 10 - 15 CERT members from each district: Divisional/District Coordinators, Staff Nuwara-eliya (rural) Exchange Situational Reports, process them to determine response resources Following a hazard event activate CERT members and EM Coordinators to respond to the event REPORTING ALERTING Two key communication functions evaluated with respect to community-based emergency response ITU-T P.800 recommended Mean Opinion Score (MOS) voice quality test for the Four Districts 51.84% Circuit (or mechanical) noise degrades the MOS with ~ 50% of the voice samples were bad, poor, and fair Such "partial", "unclear", or "missing" information can lead to false predictions or actions and inefficiencies Approximately ~30% of the voice recordings were difficult to decipher (or process) All telephone generate samples were worse or much worse than the same voice sample recorded on-site with a a digital recording Given MOS=3.52 Speaker-Independent Given MOS=3.39 Speaker-dependent Difficult 13.24% Speaker-dependent exercise with key words made is easy for the independent evaluators to predict the voice sample content (emulates a trained system) Speaker-independent, large vocabulary continuous speech, voice samples were hard for the evaluators decipher (i.e. emulates an untrained system) User variability makes a→ speaker-dependent system unfavorable (i.e. impractical to train each user) Perhaps a hybrid may be possible Emulates ITU-T P.800 recommended Percent Difficult (or Difficulty score) voice quality test for the Four Districts Difficult 29.44% telephones generated voice quailty too poor to apply "automatic speech recognition" for automating information exchage between Freedom Fone and Sahana Objectively assessed complexities of CERT members interacting with the Freedom Fone IVR in listening to Alert content and leaving-a-message of an incident report (n=47 participants); results from the District-wise controlled-exercises Extremely easy or easy 68.92% Average number of attempts made by a CERT member in listening to Alert content or leaving-a-meeage of an incident report (n=47 participants); results from the District-wise controlled-exercises Two or less attempts 86.42% CERT average response to the "Technology Acceptance Model" ease-of-use and usefulness questionnaire; conducted with n=38 participants from the four Districts With respect to their attitude towards using the Freedom Fone IVR; all things considered, the IVR was perceived to be: a good idea a benificial idea a wise idea a positive idea Easy-to-Use and Useful Freedom Fone IVR was a utility for interlinking Community Emergency Response Teams www.freedomfone.org www.sahanafoundation.org Diffusion of the research results and to lobby for resource invesments in voice-enabled emergency communication technologies was made possible through a grant from the Humanitarian Innovation Fund of the United Kingdom www.humanitarianinnovation.org The action research to study the feasibility of the innovation; i.e. the integration of the Freedom Fone IVR with the Sahana IMS, was made possible through a grant from The Kubatana Trust of Zimbabwe www.kubatana.net Our gratitude to the Lanka Jathika Shramadana Movement CERT members and EM coordinators in the four Districts and the Hazard Information Hub staff for their voluntary efforts in assisting with the innovation pilot study www.sarvodaya.org Many thanks to Freedom Fone and Sahana Software Foundation communities for their technical support in operationalizing the ICTs www.freedomfone.org www.sahanafoundation.org The Incident Command Center Staff action cycles, when interacting with the disparate Freedom Fone and Sahana systems, were not a smooth transition The shortcomings of voice-based technologies are not with the CERT members in the field It's at the incident command center with the lack self-intuitive well integrated and streamlined software tools Automation of voice-to-text or text-to-voice transformations are far fetched with the current state of natural language processing; especially for non-latin scripting languages The persistent low quality voice transmissions may improve with the role out of 3G and 4G cellular networks