Applied Speech and Audio Processing: With matlab examples
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
Infobox 4.1 Weighting
For speech and hearing purposes, voice power, background noise and other sound levels are usually measured in dBA, where the signal is A-weighted before being quantified. This is the application of a frequency weighting based upon the 40-phon equal loudness contour to the signal. Thus all frequency components in the signal are weighted so that they make a contribution to the overall figure dependent upon their perceived loudness, rather than upon their actual intensity. Although this scheme appears reasonable, when applied to speech, it takes no account of the importance of different frequencies to quality or intelligibility – only to their hearing perception. Additionally it does not account for absolute loudness, since the 40-phon curve really only describes the perception of a mythical average human to a 1 kHz signal at 40 dB SPL . More concerning is the fact that these historical and pioneering data were gathered based on single-tone signals rather than complex sounds, but are almost always applied to complex sounds. Still, in the absence of better methods, and scientific data indicating otherwise, A-weighting can be considered a reasonable approximation to many situations and applications outside the boundary of its known validity. A-weighting has been ratified by the International Standards Organisation (ISO), as have ISO B- and C-weighting based on the shapes of the 70- and 100-phon curves respectively. As an aside, dBA is commonly used for level measurements of ‘nuisance’ noise. However this only works if the audio frequencies being measured fit to the 40-phon curve. Infrasonic frequencies (2–32 Hz) and ultrasonic frequencies ( > 20 kHz) both fall almost entirely outside the measurement domain. An example is given in [3] where householders near New York’s JFK airport complained about excessive rumble from passing aeroplanes. This prompted tests for noise pollution which indicated that very little low-frequency noise was present. The rumble sound came from much higher frequency harmonics inducing a low-frequency perceived sound (we will see how this happens in Section 4.4.1). On an A-weighted scale the noise level measured would be low, whereas the sound experienced by the householders was loud. The equal loudness contours as plotted in Figure 4.2 are the result of a number of factors in the human hearing system, one of which is the filtering (frequency selectivity) introduced by the pinna: orthotelephonic gain. The frequency distribution impinging on the eardrum differs when inner-ear headphones are used as opposed to loudspeakers, as the pinna provides around 5 dB gain at 2 kHz, 10 dB gain at 4 kHz and 2 dB gain at 8 kHz [8]. The filtering effect of the pinna below 500 Hz is negligible [3]. |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling