Subject: Summary: Effect of overall level for speech in noise From: Stuart Rosen <s.rosen@xxxxxxxx> Date: Tue, 24 Apr 2012 17:38:42 +0100 List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>I asked about Enrique's Q1 in 2008 and here is how I summarised the situation then: -------- Original Message -------- I recently asked for information about studies concerning the extent to which performance for speech in noise for audiometrically normal listeners depends only on SNR, and not on overall level (at least over a fairly wide range of moderate levels). A number of correspondents pointed out that this was probably not true for fluctuating noises, so I should have specified stationary noises in my original question. Sorry! ---------------------------------------------- Trevor Agus& Jim Miller pointed me to: Hawkins, J. E., Jr., and Stevens, S. S. (1949). "The masking of pure tones and of speech by white noise". J. Acoust. Soc. Am. 22, 6-13. They determined the "level at which the listener is just able to obtain without perceptible effort the meaning of almost every sentence and phrase of the connected discourse." Once the sensation level of the white noise reached about 30 dB, the SNR was independent of overall level to noise levels as high as 90 dB SL. ---------------------------- Ben Hornsby& Niek Versfeld mentioned the work of Plomp& associates: PLOMP, R.,& MIMPEN, A. M. (1979). Speech-reception threshold for sentences as a function of age and noise level. Journal of the Acoustical Society of America, 66, 1333-1342. Here stationary speech-shaped noise was used to mask sentences, and the SRT shown to be constant for normal listeners once the noise level was high enough (essentially enough to have an effect on the speech at all). But maximum speech levels were not much more than about 60 dBA. ----------------------------- Dan Freed and Ken Grant cited 'a somewhat contrary view' from: Studebaker GA, Sherbecoe RL, McDaniel DM, Gwaltney CA (1999). "Monosyllabic word recognition at higher-than-normal speech and noise levels", JASA 105:2431-2444. Here it appears that identification of words in stationary speech-shaped noise did depend somewhat on overall level in addition to SNR, for speech levels greater than about 70 dB SPL. For example, at performance levels near 50% (oh those bloody rau!) which was at an SNR of +5 dB, performance varied from about 56 to about 28 rau for speech levels changing from 64-99 dB, so about 28% over 35 dB, or just about 1%/dB which their Table VI claims is not untypical (note however that Table VI contains at least some citations to studies with non-stationary noises, so may not be strictly comparable). -------------------------------------------------------------------- Doing a little more digging around myself: Summers& Cord (2007, JASA, Intelligibility of speech in noise at high presentation levels:Effects of hearing loss and frequency region) showed performance to deteriorate for sentences in speech-shaped noise from about 34% to about 20% as speech levels increased from 75 to 87.5 dB SPL For band-bass filtered speech ranging over a 10 dB range (55-65 dB SPL) but at constant SNR against a stationary noise, Dubno et al (2007, Ear& Hearing) found performance not to vary at a constant SNR of +19 dB. Wagener& Brand (2005, IJA) showed SRTs not to vary with noise levels varying from 45-80 dB SPL (where SRTs are typically around -6 dB) ------------------------------------------------------------------------ To summarise: For stationary noises, it appears that performance does not depend on overall level to speech levels of at least 70 dB SPL and maybe more. However, once speech levels reach about 75 dB SPL, it seems pretty clear that performance worsens. Evidently, as mentioned by Ken Grant, there is a level distortion factor which is now part of the SII standard meant to account for the decrements in performance once levels get 'too high'. What came as a surprise to me is that few studies have investigated this issue carefully, particularly as regards the 'cutoff' level at which overall level starts to matter. It is also very difficult to compare results across studies because of differences in materials (which may or may not influence results) and measurement methods. Thank you all for your comments! Yours - Stuart -- /*------------------------------------------------*/ Stuart Rosen, PhD Professor of Speech and Hearing Science Speech, Hearing and Phonetic Sciences Division of Psychology& Language Sciences, UCL 2 Wakefield Street London WC1N 1PF England Tel: (+ 44 [0]20) 7679 4077 Admin: (+ 44 [0]20) 7679 4050 Fax: (+ 44 [0]20) 7679 4010 Email: stuart@xxxxxxxx Home page: http://www.phon.ucl.ac.uk/home/stuart /*------------------------------------------------*/