Real-time speech acquisition, compression and wireless transmission solution on resource-constrained embedded systems

Abstract

Resource-constrained embedded systems are electronic systems designed to perform specific tasks with minimal hardware and software resources. They are very popular and essential to building a compact and efficient system at a low cost. This paper presents an embedded system architecture for real-time acquisition and compression, utilising wireless transmission, for intelligent embedded devices. The platform uses an STM32F411CEU6 (ARM Cortex–M4) microcontroller, paired with an INMP441 MEMS microphone, and employs the Codec2 encoder at a rate of 3.2 kbps. An optimised algorithm based on receiver-side data and sending voice frame processing on I2S and UART interfaces, respectively, has been applied using CMSIS-DSP acceleration and computational constrained STM32F4 series and NRF24L01 modules, with COBS encoding. System operation results in real-time execution with a latency of 2.31 ms/ frame and a low power consumption of 50.23-51.7 mW at 3.3 V operation, demonstrating a good model with performance characteristics that simultaneously achieve minimal real-time transmission and low power consumption. The proposed architecture system is well-suited and potentially suitable for next-generation speech-centric applications such as responsive speech-to-text, real-time command recognition, and a compact on-device language translation module.