Search

Most of the noise in speech communication lines can be considered as Gaussian white noise. Voice activity detection (VAD) in noisy environment is an important process in many speech signal processing algorithms. Unlike the other VAD algorithms, this paper proposes a simple and novel VAD algorithm based on the probability distribution function (PDF) of FFT magnitudes of both clean speech and Gaussian white noise. When the signal-to-noise ratio (SNR) is high enough, the method using Gamma distribution to detect the speech performs well, while the method using Rayleigh distribution under lower SNR can be complementary. In addition, the threshold to determine which method to use is presented based on the tests under different SNR. Simulation results show that the proposed algorithm is efficient. Both the hardware and software of a low cost system for VAD are introduced, with the proposed algorithm achieved in a digital signal processor (DSP). Each detection takes on less than 100 ms, which can be used for real-time processing.