Increasing robustness in the calculation of the speech transmission index from impulse responses

There are many factors that can affect the measured values of the speech transmission index (STI), and this paper examines how and why identical inputs into STI calculation software are yielding varying results. The study involved a survey of a number of software implementations of the Indirect Method for computing the STI from an impulse response, one of which was written by the authors. Results are presented for artificial and measured impulse responses, and for signal and noise spectra that were designed to test particular aspects of the STI calculation. While most deviations between implementations were within 0.01 STI, some were not, revealing a need for greater robustness in the design of software and greater clarity in the STI standard (IEC60268-16), including more support for validation. This paper provides some data for such validation