In this letter, we develop a robust voice activity detector (VAD) for the a
pplication to variable-rate speech coding. The developed VAD employs the de
cision-directed parameter estimation method for the likelihood ratio test.
In addition, we propose an effective hang-over scheme which considers the p
revious observations by a first-order Markov process modeling of speech occ
urrences. According to our simulation results, the proposed VAD shows signi
ficantly better performances than the G.729B VAD in low signal-to-noise rat
io (SNR) and vehicular noise environments.