Speech Recognition System Using DeepSpeech Architecture Method on VHF Radio Communication for Tanker Ship Officers at Indonesian Sea Ports

Abstract

The loading and unloading of fuel oil cargo by tanker ships at ports in Indonesia has a problem in terms of time efficiency and speed. A virtual robotic is created to increase the time efficiency of the loading and unloading process. However, the robot needs a way to communicate with the tanker ship officers during the process. Because the port where the loading and unloading process took place is considered as a dangerous and explosive location, the only communication allowed is through voice communication via marine Very High Frequency (VHF) radio. The solution to overcome this problem is to design a technology that can perform speech recognition via marine VHF radio, one of which is using the Deep Learning method with DeepSpeech architecture. This paper has simulated speech recognition system using DeepSpeeh architecture method on VHF radio communication for tanker ship officers at sea ports. This paper has tested the DeepSpeech architecture to produce a speech recognition model with an average Word Error Rate (WER) value of 0.335 and an average Character Error Rate (CER) value of 0.263. This paper also analyzes the effect of variations in learning rate, dropout rate, and epoch value to get the best speech recognition system model.