The accuracy of noise estimation is important for the performance of a speech enhancement system. This study proposes using variable segment length for noise tracking and variable thresholds for the determination of speech-presence probability. Initially, the fundamental frequency is estimated to determine whether a frame is a vowel. In the case of a vowel frame, the segment length increases; meanwhile the threshold for speech-presence is decreased. So the noise magnitude is adequately underestimated. The speech distortion is accordingly reduced in enhanced speech. Conversely, the segment length is rapidly decreased during noise-dominant regions. This enables the noise estimate to be updated quickly and the noise variation to be well tracked, yielding background noise being efficiently removed by the process of speech enhancement. Experimental results show that the proposed method can efficiently track the variation of background noise, enabling the performance of speech enhancement to be improved.