LAUSR

This letter introduces a method based on the Hilbert–Huang transform (HHT) to estimate the fundamental frequency of nonstationary noisy speech signals. For this purpose, the target signals are analyzed by means of the ensemble empirical mode decomposition and the Hilbert transform. The main contribution of the proposed solution, namely HHT-Amp, relies on the extraction of pitch information from the instantaneous amplitude of the first decomposition modes. The HHT-Amp and four competitive algorithms are evaluated considering speech signals corrupted by five acoustic noises with different nonstationarity degrees. The HHT-Amp achieves the lowest gross error rate and mean absolute error for the most severe noisy conditions. This demonstrates that the proposed approach outperforms the baseline methods in estimating the fundamental frequency of noisy speech.