Intelligibility prediction for speech mixed with white Gaussian noise at low signal-to-noise ratios

The effect of additive white Gaussian noise and high-pass filtering on speech intelligibility at signal-to-noise ratios (SNRs) from −26 to 0 dB was evaluated using British English talkers normal hearing listeners. SNRs below −10 were considered as they are relevant security applications. Eight objective metrics assessed: short-time (STOI), a proposed variant termed STOI+, extended (ESTOI), normalised covariance metric (NCM), subband envelope correlation (NSEC), two derived the coherence index (CSII), an envelope-based regression method transmission (STI). For mixtures associated with scores ranging 0% 98%, STOI+ performed least well other and, under some conditions, better than STOI, ESTOI, STI, NSEC, CSIIMid, CSIIHigh. Both NCM relatively low prediction error bias for dB. STI in terms scores, error, bias, reliability. Logistic modeling demonstrated that filtering, which increases proportion high frequency energy, detrimental between −5 −17 inclusive.