Compensating the STI to predict intelligibility for amplitude-compressed speech

The speech transmission index (STI) has not been able to quantify speech intelligibility for amplitude-compressed speech even though it is widely accepted as a successful physical metric to predict speech intelligibility for normal hearing listeners in additive noise and reverberation conditions. This paper presents a modification to a speech-based STI such that it can be used to predict the intelligibility for hearing impaired listeners using amplitude-compression hearing aids. By analyzing the effect of slow-acting wide dynamic range compression (WDRC) on sinusoidally modulated envelopes, a transformation that relates the compression ratio to the expected modulation reduction is determined. Based on this transformation, a modification to the STI is proposed that provides consistent STI values for both amplitude-compressed and linearly amplified hearing-aid processed conditions when they have the same intelligibility. We then propose a model based on the linear regression technique. We show that, with such a model, the speech intelligibility for sensorineural hearing-impaired listeners can be predicted using the modified speech-based STI