Voice Activity Detection in Noisy Environments Based on Double-Combined Fourier Transform and Line Fitting
Figure 3
Results of endpoint detection based on frame energy and employing an edge detection filter in a car-noise environment: (a) noisy speech signal at SNR −5 dB, (b) clean speech signal, (c) frame energy of signal (a), and (d) detection output of edge filter with state transition model ( is lower threshold; is upper threshold).