The sound level of each of 16 diverse noises was adjusted so that loudspeaker‐reproduced rhyme words at a level of 78 dB (C) at one meter were reduced to 50% word intelligibility. Spectrum analyses (octave band and 20‐cps band), physical measurements (C, B, A, DIN 3 sound‐level meter weightings), calculations [Speech Interference Levels (SIL)], and peak fittings [to noise criteria (NC)‐type curves] were made to find which method came closest to agreeing that the noises were equally speech‐interfering.
SIL calculations that included the octave 300–600 cps predicted the speech‐interference aspects of the 16 noises best. The A and DIN 3 weighting networks, the conventional (600–4800 cps) SIL, and use of a restricted region (350–2800 cps) of NC‐type curves gave the next best prediction. The poorest predictors were the B or C frequency‐weighting networks and use of the NCA, NC, or ISO contours over their total frequency range.