Intelligibility and listener preference of telephone speech in the presence of babble noise
Spectral modification introduced by differentiation (solid line) and by formant equalization (dashed line). z-transforms used for differentiation and for formant equalization are given in the Appendix. The figure also shows the handset “send” (dash-dot line) and “receive” (dash-triple dot line) frequency response masks for wideband telephone (TIA/EIA-920, 2002).
Average spectrum of speech babble used as masker (solid line). The babble shows a high-frequency noise floor about 25 dB below the peak at 2 kHz. Average spectra for unprocessed speech from the two talkers used in experiment 1 (dashed lines). Note that the spectrum for babble, representing masking generated locally, is band limited to 8 kHz, while spectra for the two talkers, representing speech received over a wideband telephone system, are band limited to 7 kHz.
Intelligibility (probability of a correct response) versus spectral conditioning for the average of the two talkers (AV), for the female talker (F), and for the male talker (M). UN: unprocessed. FE: formant equalization. DIFF: differentiation. Results averaged over four experimental subjects. Vertical bars give 90% confidence intervals. Babble masker at 70 dB SPL and speech at −6 dB SNR.
Listener preference for speech processed by formant equalization over unprocessed speech (dashed line), speech processed by differentiation over unprocessed speech (dash-dot line), and speech processed by formant equalization over speech processed by differentiation (solid line). Babble masker at 70 dB SPL.
Similar to Fig. 4 except that results are shown for speech rich in glides and nasals (upper panel) and for speech rich in fricatives (lower panel). Babble masker at 70 dB SPL.
Article metrics loading...
Full text loading...