Auditory modeling applied to formant tracking of noise-corrupted speech

The authors explore the use of the EIH (ensemble interval histogram) as the basis for formant frequency estimation for noise-corrupted speech. By measuring the percent RMS (root mean square) errors of the first three formants it was shown that the EIH is effective for formant frequency estimation. The authors also show the effects that some of the model parameters have on formant tracking and conclude that the uniform IIR (infinite impulse response) filter bank with eight linear level-crossings is best suited for formant tracking of noise-corrupted speech. The minimal influence of these parameter changes indicates the robustness of the auditory model. Results obtained provide strong evidence that formant frequency estimation based on the EIH may be more accurate for noise-corrupted speech than results achieved with traditional methods