Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
A study of the application of automatic speech recognition in assistive robotics
Submission Author:
Walter de Britto Vidal Filho , DF , Brazil
Co-Authors:
Pedro Pereira Nunes, Walter de Britto Vidal Filho
Presenter: Walter de Britto Vidal Filho
doi://10.26678/ABCM.COBEM2023.COB2023-1007
Abstract
Technologies that allow automatically converting speech into text are strongly disseminated in today's society. A common application is to provide for visually impaired people the ability to interact with a cell phone more easily, providing its inclusion. Other possibilities to positively impact lives of people with disabilities were investigated with the aim to integrate automatic speech recognition technology with robotics, in order to contribute to the tasks performance for people with limitations in their motor functions, especially limited in the use of their arms and hands. This research sought to implement a voice command interface for a robotic arm. To achieve this were analyzed the most appropriate words to compose the commands of this interface, based on ease speech and understanding, and sequentially, habilitate the use of the commands to control the position of the end effector. The research consisted in creating the types of commands to be implemented, including two main types: commands that indicate the coordinate where the robotic manipulator should stop and commands that do not indicate it and therefore needed a stop command. In order to validate the set of commands, acceptance metrics for the words that compose the lexicon were determined, such as false-positives and false-negatives. Additionally movement algorithms which respond to voice commands were implemented to verify the usability of the robotic arm for moving objects in its environment. Usability was measured based on the time and number of commands needed to perform a task according to the two main types of commands. More than 30 words selected from Brazilian habitual speech were examined to compose the interface and command a SCARA-type handler using a voice recognition API. The chosen words had an error rate of less than 10%. About 25 commands were implemented to move the arm along three-dimensional space and to control the end effector. It was observed that the set of commands that have prior knowledge of target coordinates makes the voice interface easier to operate in terms of task execution time and practicality, considering that stopping commands given by the user need more attention related to arm’s path. According to the results obtained, an interface that uses only voice to control the robotic arm can be integrated, especially if the target coordinates are known.
Keywords
Robotics, assistive technology, speech recognition
