Real-Time Text and Language Independent Speaker Identification with a Reconfigurable Wireless Network of Acoustic Sensors

This paper describes a reconfigurable wireless network of acoustic sensors that records voice signals in different areas of a building and conveys them at the sink node. At their arrival, a light-weight text and language independent algorithm performs the speaker identification task in real time. The end-user can interrupt the normal operation mode of the network and require a signal to a particular node, specifying also sampling frequency and time length of the sampling period. In our simulations, we use a database composed of 200 signals, 60 individuals, and 15 languages. The total execution time is less than 2 seconds. We optimize the parameters of the algorithm, achieving 83% accuracy. We also evaluate its robustness when the sampling frequency and the time length of the signals are reduced. Finally, the power consumption of the operating nodes is analyzed.