Speech Recognition System Using Open-Source Speech Engine for Indian Names

摘要

Speaker independence, continuous speech and huge vocabularies create most of the greatest challenges in automatic speech recognition. This paper describes Sphinx, a library that offers the feasibility of accurate, huge vocabulary, speaker-independent, continuous speech recognition. Using speech for device control is a proven hands-free solution. There are several products that use speech input for hands-free control. They usually cater the users with US/UK accent. In this paper, a speech recognition system is developed for the application of hands-free control system to be deployed in automotive environment for Indian users. This paper demonstrates the methodology and the challenges of customizing an open-source speech recognition engine for Indian users. It is demonstrated for the application of speech-based control of smartphone, and rear-view mirror rotation. Open-source package used is Pocketsphinx for speech recognition and festival for text-to-speech and pronunciation generation. All the implementations are done on a single-board computer, i.e. raspberry pi.