Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Gesture recognition for sentence synthesis with probabilistic algorithms

Giannelos Evangelos

Full record


URI: http://purl.tuc.gr/dl/dias/7123ABC9-7723-4E8F-9BB0-61516EDCBA6A
Year 2018
Type of Item Diploma Work
License
Details
Bibliographic Citation Evangelos Giannelos, "Gesture recognition for sentence synthesis with probabilistic algorithms", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2018 https://doi.org/10.26233/heallink.tuc.78338
Appears in Collections

Summary

The purpose of this work was to develop a standalone system that predictstext and outputs speech synthesis for people with physical disabilities andreduced fine motor function. A low-cost, infrared hand tracking device wasutilized, typically used in virtual/augmented reality applications. The implementedsystem offers two basic functionalities. First, a human-computerinteraction (HCI) functionality that is customizable and offers enough featuresthat can be tailored to a large number of different users. Second, a wayto predict what the user intends to write in order to reduce the user’s inputtext as much as possible, speeding up overall communication. The wordprediction task was mainly based on n-grams (and relevant Markov Chains),abbreviation expansion (and lookup tables) and recurrent neural networks.Evaluation results confirmed that the simple prediction methods utilized acceleratedword typing speed, saving input text by approximately 63%. Futurework could utilize additional eye-tracking or hand-tracking sensors.

Available Files

Services

Statistics