Time-Series Analysis for Ear-Related and Psychoacoustic Metrics V. Mellert, H. Remmers, R. Weber, B. Schulte-Fortkamp
how to analyse p(t) to obtain an earrelated parameter? general remarks on acoustical analysis ear-related signal processing hearing sensation sensation of vibration time-frequency representations
acoustical analysis time related level, envelope statistical level-/ envelope-distributions modulation-frequency analysis estimates for impulsive sound
acoustical analysis related to spectral properties weighting functions (A, B, C,...) octave, third-octave spectrum high resolution Fourier analysis bandwidth time-window 1
ear-related signal processing two ears spatial hearing source localisation suppression of disturbing noise interaural correlation Measuring system Head Acoustics
ear-related frequency analysis p(t) head diffraction ear canal Impedance - transformation Basilar membrane Cochlear micromechanics excitation at low frequencies excitation at high frequencies
ear-related frequency analysis p(t) head diffraction ear canal Impedance - transformation Basilar membrane Cochlear micromechanics movie from Dept. of Neurophysiology, Univ. of Wisconsin Madison http://www.neurophys.wisc.edu/h%26b/auditory/animation/animationidx.html
frequency location low frequencies time time dependant excitation pattern location spatial exitation --------- Basilar membrane resp. equivalent set of band-pass filters f(t) high frequencies
temporal integration nonlinear loudness 100 200 ms masking effects : Cited from Zwicker/ Fastl: Psychoacoustics Facts and Models
frequency (spatial) integration loudness: appr. 1/3 octave (bark-scale) modulation synchrony (object forming, binding problem) Cited from Zwicker/ Fastl: Psychoacoustics Facts and Models
loudness summation scheme [Zwicker] p(t) bandpass filters: 1/3 ocatave, gamma-tone, ftt [Terhardt] envelope low pass integration nonlinearity envelope low pass integration nonlinearity masking properties sum N(t)
hearing sensations loudness [Zwicker] sharpness, sensory pleasantness [v. Bismark, Aures, Terhardt]
hearing sensations loudness [Zwicker] sharpness, sensory pleasantness [v. Bismark, Aures, Terhardt] roughness [Aures,Terhard, Daniel] fluctuation strength [Fastl]
hearing sensations loudness [Zwicker] sharpness, sensory pleasantness [v. Bismark, Aures, Terhardt] roughness [Aures,Terhard, Daniel] fluctuation strength [Fastl] algorithms for measurements partially available
hearing sensations loudness [Zwicker] sharpness, sensory pleasantness [v. Bismark, Aures, Terhardt] roughness [Aures,Terhard, Daniel] fluctuation strength [Fastl] algorithms for measurements partially available Comprehensive model of the functionality of the ear: effective signal processing [Dau, Kollmeier]
sensation of vibration: thresholds 140 db 1 g
threshold measurements: z
threshold measurements: y
vibration total value after ISO 2631-1/2 a V = [ ] 1 nk na nk na nk na k a + k a k a na x W, x y W, y + z W, z
vibration total value after ISO 2631-1/2 a V = [ ] 1 nk na nk na nk na k a + k a k a na x W, x y W, y + z W, z spatial weights e.g. 0,..., 1
vibration total value after ISO 2631-1/2 a V = [ ] 1 nk na nk na nk na k a + k a k a na x W, x y W, y + z W, z frequency weighted vibration rms-values for x, y, z direction
vibration total value after ISO 2631-1/2 a V = [ ] 1 nk na nk na nk na k a + k a k a na x W, x y W, y + z W, z a = T 0 T 1 2 W, x / y ( asx / y( t)) a W T 1, = z ( asz( t)) T 0 2 dt dt 1 2 1 2
Just-noticable vibration difference value
multispectrum frequency resolution 1/3 octave, except for frequencies below 200 Hz time resolution, signal duration loudness fluctuation: > 1 s stationary: at least 200 ms fluctuation strength: time window < 50 ms roughness: time window < 5 ms
interior noise of a passenger jet seat #4, 200 ms length, relative level in steps of 2.5 db 0 29 58 87 116 145 174 time [ms] 5k 4k 3.15k 2.5k 2k 1.6k 1.25k 1k 800 630 500 400 315 200-250 100-160 16-80 frequency [Hz] 0.0-2.5 2.5-5.0 5.0-7.5 7.5-10.0 10.0-12.5 12.5-15.0 15.0-17.5 17.5-20.0
interior noise of a passenger jet seat #5, 200 ms length, relative level in steps of 2.5 db 0 29 58 87 116 145 174 time [ms] 5k 4k 3.15k 2.5k 2k 1.6k 1.25k 1k 800 630 500 400 315 200-250 100-160 16-80 frequency [Hz] 0.0-2.5 2.5-5.0 5.0-7.5 7.5-10.0 10.0-12.5 12.5-15.0 15.0-17.5 17.5-20.0
technical multispectrum linear transformations (log. amplitude) seven 1/3 octaves below 80 Hz three 1/3 octaves from 100 to 160 Hz two 1/3 octaves from 200 to 250 Hz additional tracks of weighted vibration signal time window 6 ms signal duration 200 ms
towards an ear-related representation keeping time frequency resolution including masking effects (loudness meter) object binding (identification of tonal components)
part-tone time-pattern pttp [Terhardt, Heinbach, Mummert, 1985 1998] location time time dependant excitation pattern frequency contour-plot: line-pattern of significant features time
4 th order Fourier-time-transformation FTT [Terhardt] (gamma-tone, wavelet) identifiy local maxima (level 3 db) in cuts parallel to frequency axis identify local changes in level with respect to time in cuts parallel to time axis connect identified points to lines according to threshold criteria Ommit short lines (texture) and keep long ones (contour)
pttp-contour of a tyre noise
high-resolution pttp-contour of a helicopter in flight
multispectrum of a helicopter in flight 0 29 58 87 116 145 174 time [ms] 5k 4k 3.15k 2.5k 2k 1.6k 1.25k 1k 800 630 500 400 315 200-250 100-160 16-80 frequency [Hz] 0.0-2.5 2.5-5.0 5.0-7.5 7.5-10.0 10.0-12.5 12.5-15.0 15.0-17.5 17.5-20.0
technical realisation/ measurements no spatial information: mono microphone acceleration in z sufficient (sensitivity) single value for vibration in multispectrum 200 ms time history (steady state) time window for spectrum > 5 ms no masking
summary and future developement time-frequency representations are similar to the excitation pattern of the ear data reduction schemes like pttp can improve the multispectrum representation additional spectral analysis of the envelope provides effective information about time fluctuations the images are an efficient input for pattern recognition schemes (like ANN)