ABSTRACT: Ever want a robot to reciprocate your feelings about what you are thinking about? What if it is going onto your favorite Star Wars bot? Well, don't look any further. It is right here (just in development, of course).
PROJECT DEVELOPMENT:
-
Speech Recognition: In the Fall 2018 semester, the development of a minimum viable product (MVP) has been completed with a Raspberry PI and a ReSpeaker Microphone. This MVP is equipped with a simple NLTK sentiment analysis algorithm that uses a bag of words model to create a numeric representation of what a user is saying between -1 and 1 (negative to positive). Some issues of this model is that it only takes into account the individual words in the sentence, not the connotation of the sentence itself. For example, negation statements such as "not" are not interpreted correctly in statements that, when analyzed, seem negative but are actually positive.
-
Sentiment Analysis Algorithm: In the Spring 2019 semester, the Sentiment Analysis API has been switched to IBM's Watson instead of the NLTK API. The same supplies are being used. The codebase has been updated to account for a number of presentation cases that will be demonstrated during the Cornell Cup competition, as the voice-based system that has been created is going to be used as the primary user interface. Updates to the microphone were implemented as well to assist other projects in obtaining direction of arrival and of decreasing the amount of background noise the micophone picks up.
SOURCES USED: http://wiki.seeedstudio.com/ReSpeaker_Mic_Array_v2.0/ : microphone used in creating the MVP --- https://github.com/respeaker/respeaker_python_library : used for speech-to-text implementation --- https://www.nltk.org/ : used for basic sentiment analysis