Small Vocabulary Speech Recognizer

Overview

• Project for IF4071 Speech Processing course on Bandung Institute of Technology.
• Develop an Automatic Speech Recognition system for Indonesian Language small vocabulary by employing one of the most popular tool.
• Use HTK Tool and Julius
• Submitted on November 29th, 2020

Project Milestone

• Tools installation and exploration (1 week, Oct 26 - Nov 8)
• Data preparation (1 week, Nov 8 - Nov 15)
• Experiments and documentation (2 weeks, Nov 15 – Nov 29)

Development Details

1. Tools Exploration (Reading Material: HTK Book, Julius)
2. Preparation of speech corpus:
   • Audio file recorded from all group members spoken vocabulary/sentences of selected domains in Indonesian Language. (E.g. For digit recognizer satu, dua, tiga, ...sepuluh, ...)
   • Recorded in a clean environment (silent condition)
   • Use high-quality microphone
   • Divide speech corpus into training and test corpus with ratio 9:1.
3. Build an ASR for recognizing small vocabs (Chapter 3: A Tutorial Example of Using HTK)
4. ASR Evaluation using:
   • Closed experimental condition : use training corpus as evaluation corpus
   • Open experimental condition: use test corpus as evaluation corpus