Abstract: In order to detect voiced Chinese speech especially in various noise environments at low SNR (signal-to- noise ratio), a robust voiced speech detection method was proposed by the harmonic structure of voiced speech. A set of lines which represent the harmonic structure of voiced speech were tracked by improved spectral tracking method, and characters were picked up as foundation to voiced Chinese speech detection from the spectral lines. It's better than parallel methods proved in the comparison experiments which test data obtained in different SNR and various noise environments. Especially when the SNR is 0 dB, its accuracy rate is nearly 30 percent higher than parallel methods. Experiments demonstrate the effectiveness of the voiced Chinese speech detection method in various noise environments at low SNR.