Underwater Diver Communication System Based on Speech/Text Conversion

Underwater acoustic (UA) communication system has low data rate due to the limited bandwidth of the UA channel. This makes real-time speech communication challenging. In this paper, we present an underwater diver communication system which uses speech-to-text conversion at the transmitter to achieve speech compression. The text message is converted back to speech at the receiver end using text-to-speech conversion. Automatic speech recognition (ASR) models including Wav2vec2 and Whisper are investigated for their feasibility of real-time conversion. A prototype underwater diver speech communication system utilizing Raspberry Pi 5 is presented in the paper. The system design encompasses both the transmitting and receiving processes, with a focus on its real-time capabilities. We discuss the performance of this diver communication system, validating its efficacy through an experiment conducted in Fremantle, Western Australia. It is shown that with the Whisper-base model, real-time communication is achieved with an average word error rate of 2.4%.