Comparison of speech enhancement algorithms for forensic applications
Ahmed Kamil Hasan Al-Ali, David B. Dean, Bouchra Senadji, Vinod Chandran · QUT ePrints (Queensland University of Technology) · 2016
Speech enhancement algorithms play an essential role in forensic applications, and enhanced speech signals can be used in court as evidence in criminal cases. This paper compares the performance of single channel (spectral subtraction and level dependent wavelet threshold techniques) and multiple channel (independent component analysis or ICA) speech enhancement algorithms to remove real environmental noise from noisy audio recording signals. Experimental results demonstrate that ICA achieves a significant improvement in average signal to noise ratio (SNR) enhancement compared to single channel speech enhancement algorithms, when 100 sentences from a forensic voice comparison database were corrupted with a car, street and factory noise at input SNR (-10 to 10 dB).