Implementation of Python-Based Korean Speech Generation Service with Tacotron

Abstract

This paper deals with the implementation of the Korean speech generation service based on Python. Speech generation technology is used in various kinds of information and communication services and has been further refined by the development of deep learning technology. Tacotron, a representative deep learning speech generation model, is implemented on the server, and CherryPy is used to provide speech generation samples for queries through a web browser. The model parameters obtained through pre-training are stored and generate the voice as soon as the Korean query is received from the web server. This model can be extended and used in chatbot systems and voice guidance services.