Syntax Independent Code Generation Method and Compiler Device from Voice Input based Template Grammar Specification using Machine Learning

Drake A. Johnson
Dr. Kafi Rahman, Faculty Mentor

Code generation can reduce repetitive coding and automate the creation of common application source code for any language [1]. In this research, we present a framework to translate a vocal specification of a program to source code for a target computer. By using speech recognition and neural networks (NNs), we generate program code from processing commands such as, “Create a function that returns void and takes no parameters.” Commands like these are then processed to text, passed into the input layer of a NN, and matched with a set of grammar-driven [2], parse-tree-like templates for a specified programming language. Breaking this code generation process into several phases makes it easier to understand and allows for improved quality of generated code. In this work, architecture and system details of the proposed framework are described. Currently, the framework supports code generation in C++ and Python programming languages.

 

Keywords: Machine learning, Neural network, Voice recognition, Speech processing, Source code generation, Natural language

Topic(s):Computer Science

Presentation Type: Oral Presentation

Session: TBA
Location: TBA
Time: TBA