Goals
Developed between 1964-1966, Eliza was the early natural language processing computer program, which had a goal to demonstrate superficial communication between man and machine. Created by MIT Artificial Intelligence Laboratory by Joseph Weizenbaum, created out of frustration that the Artificial intelligence methods by that time did not produce satisfying progress. His goal was to create an application that simulated, famous by the time, method of psychoanalysis where the doctor repeats information from the subjects in the form of questions. The results were surprising for Weizenbaum too and the program become one of the first examples of capable to attempt the Turning test.
Our attempt to modernise this already impressive old system was to transform it to the digital domain and current web technology. We used a few APIs from Google and Web Speech API to create a system that can, record, recognise and synthesise voice in order to create a voice-enabled experience. We further tried to detect the mood of the speaker and adjust the prosodic features of the synthesised voice to match the mood. For example, a sad person, should not receive a happy voice when talking about his or her problems.
Technology and Implementation
This project was developed with backend python and python web application, which communicates between a webpage and an instance of the traditional Elisa program but implemented on python. We used Web Speech API to detect the sound input and translate it to text, after which we used TextBlob's Sentiment Analysis python library to find a numerical representation of the mood. This input is processed by our Dialogue management system and finally, Google's API text-to-speech is used, where we adjust prosodic properties to reflect what we detected from the input.
The whole application was created for the experiment, where each visitor to the website was randomly assigned to the control or experimental group and statistical data was collected for the experiment at hand. The system was deployed on Google Cloud Service.