Improving syllable identification by a preprocessing method reducing overlap-masking in reverberant environments.
Speech intelligibility in reverberant environments decreases due to overlap-masking. Unlike additive noise, the masking signal is not independent from the information bearing signal. A mathematical framework for intelligibility-enhancing signal modification prior to presentation in reverberant environments is presented in this paper. The optimal solution generalizes steady state suppression and adjusts the short-term signal power as a function of late reverberation power and signal importance. The signal modification operates in a full-band setting and preserves the time scale of the unmodified signal. Gain smoothing based on an adaptive rate-of-change constraint reduces processing artifacts and enhances performance. Subjective validation shows that the proposed method effectively reduces the impact of overlap-masking. Speech intelligibility at a reverberation time of 1.8 s was improved significantly compared to unmodified and steady-state-suppressed speech.