Comparison of approaches to estimate the speech modulation transfer function

Using speech as a probe stimulus to compute the Speech Transmission Index (STI) has been of great interest to speech researchers. One technique is based on first computing the speech modulation transfer function (SMTF). Approaches used to obtain the SMTF include those developed by Steeneken and Houtgast [H. Steeneken and T. Houtgast, Proc. 11th ICA, Paris 7, 85–88 (1983)] and Drullman et al. [R. Drullman, J. M. Festen, and R. Plomp, J. Acoust. Soc. Am. 95, 2670–2680 (1994)]. This paper compares these two approaches and a new one to the theoretically obtained modulation transfer function (MTF) for reverberant and noisy environments. The new method computes the magnitude of the cross‐power spectrum rather than the real part used by Drullman. As previously reported, Houtgast’s method exhibits artifacts at high modulation frequencies. Drullman’s approach eliminates artifacts in the reverberant environment but does not predict the theoretical MTF for the noisy environment. The new method outperforms the other two approaches in matching the theoretically derived MTF across both environments. This paper also examines the SMTF of amplitude‐compressed speech for these three methods. [Work supported by NIDCD.]