Human Psychoacoustics

Transcription

1 1 ESE250: Digital Audi Basics Week 6 February 19, 2013 Human Psychacustics

2 Curse Map 2

3 Week 2 Received signal is sampled & quantized q = PCM[ r ] Week 4 Sampled signal first transfrmed int frequency dmain Q = DFT[ q ] Week 3 Quantized Signal is Cded c =cde[ q ] Week 5 signal versampled & lw pass filtered Q = LPF[ DFT(q+n) ] Week 6 Transfrmed signal analyzed Using human psychacustic mdels Week 7 Acustically Interesting signal is perceptually cded C = MP3[ Q] Where are we? r(t) p(t) Prduce Week 4 Week 6 Over Sample Decde q + n DFT Stre / Transmit Q + N C Week 5 Week 3 LPF Q Perceptual Cding [Painter & Spanias. Prc.IEEE, 88(4): , 2000] 3

4 The Physical Ear [R. Munkng and B.-H. Juang. IEEE Sig. Prc. Mag., 25(3):98 117, 2008] External Sund Waves Guided by uter ear int auditry canal Excite Inner Ear Thrugh mechanical linkage cnnecting ear drum t cchlea 4

5 The Physical Ear [R. Munkng and B.-H. Juang. IEEE Sig. Prc. Mag., 25(3):98 117, 2008] Initiates signal prcessing frequency dmain analysis Via analg cmputatin Vide: Cchlea What part f the Cchlea vibrates fr an 800 Hz square wave? 5

6 Mdern Psychacustics Benefits greatly frm The Cgnitive Ear decades f neural recrding cntemprary brain imaging technlgy [R. Munkng and B.-H. Juang. IEEE Sig. Prc. Mag., 25(3):98 117, 2008] 6

7 Pwer Spectrum Mdel f Hearing B.C.J. Mre. Int.Rev.Neurbil., 70:49 86, Rugh Picture (main cntent f tday s lecture): Critical Bands: Auditry system cntains finite array f adaptively tunable, verlapping bandpass filters Frequency Bins: humans prcess a signal s cmpnent (against nisy backgrund) in the ne filter with clsest center frequency Masking: certain signal cmpnents in a given band are favred and thers are filtered ut Established thrugh decades f psychacustic experiments 7

8 Auditry Threshlds In the lab, yu varied the frequency, amplitude and phase f signals What was the effect f each, if any, n the sund yu heard? Frequency Amplitude Phase s( t) Asin(2 ft ) 8

9 Auditry Threshlds Harvey Fletcher (1940) Played pure tnes varying frequency, f [ Hz] Intensity, I [Dyn cm -2 ] = 10-5 [N cm -2 ] = 0.1 Pa phase changes tend t be inaudible Large listener ppulatin Yung Acute Recrded extreme threshlds faintest audible greatest tlerable (

10 Auditry Threshlds Results: [H. Fletcher. Rev. Md. Phys., 12(1):47 65, 1940]. pain-free hearing range extends at mst ver 20 Hz 20 KHz with sensitivity» Pa = 20 Pa 0.1 Pa 10

11 The decibel unit Define standard pressure: p 0 = Pa = 20 Pa Threshld f human hearing Cmpute Sund Pressure Level as: L SPL = 20 lg 10 (p/p 0 ) db L SPL fr p 1 = 20 Pa, fr p 2 = 200 Pa, fr p 3 = 20 mpa Cmpare t Ambient sea-level pressure: 1 Atmsphere = 10 5 Pascal 0.1 Pa Q: why use lg-lg scale? A 1 : dynamic range A 2 : ludness is a pwer functin 11

12 The decibel unit Hearing intensity ( 12

13 Let s try t reprduce these results! We will listen t single sine tnes starting at a frequency f 10KHz, all the way up t 20KHz, s each student can figure ut their cut-ff frequency Suggestins t imprve this experiment? ( 13

14 Animal hearing ranges Dgs: Greater hearing range: 40Hz t 60KHz Ultrasnic dg whistles Mice: Large ears in cmparisn t their bdies Hearing range: 1KHz t 70KHz Can t hear lw frequency nises Cmmunicate with high frequency Distress call (40KHz), alert f predatr [Pictures frm Wikipedia] 14

15 Why Sinusids? Why nt sme ther harmnic series? Furier s analysis shws harmnic analysis culd be based n arbitrary smth peridic fundamental Why des the animal receiver use sinusids? Hamiltnian Mechanics Simplest physical mdel f vibrating masses Cupled spring-mass-damper mechanics Prduce sinusidal harmnics Vide: Cchlea b. all sund is prduced by vibrating masses. m x k 15

16 Masking - Spatial Masking Paradigms Masker masking maskee Tne Masking Nise pure tne f 80 SPL at 1 khz just masks critical band nise f 56 SPL centered at 1 khz Masker-t-Maskee rati Cnstant fr fixed relative frequency and varying amplitude Changes with varying relative frequency 1 Bark frequency interval [T. Painter and A. Spanias. Prc. IEEE, 88(4): , 2000.] 16

17 Masking The first graph shws the masking pattern fr a 200Hz tne Mstly masks tnes arund 200Hz, but als at harmnics The secnd graph shws the same plt fr different frequencies, but nly the fundamental part Ntice that the band gets wider fr increasing frequencies masker at fundamental can smewhat mask maskees at the harmnics [H. Fletcher. Rev. Md. Phys., 12(1):47 65, 1940]. but the spreading curve is traditinally depicted ver the fundamental nly 17

18 Tne Masking Nise Are the fllwing signals masked? 200 Hz tne at 80dB 200 Hz tne at 40dB 300 Hz tne at 40dB 400 Hz tne at 40dB 700 Hz tne at 30dB 18

19 Masking [H. Fletcher. Rev. Md. Phys., 12(1):47 65, 1940]. Tne Masking Nise (Fig 12) value abve quiet threshld such that a signal at the abscissa frequency can be heard in presence f tp: 200 Hz tne bttm: varius frequencies Nise Masking Tne (Fig 13) dts shw pure tne magnitude (in db) required t be audible abve nise Of the magnitude n the middle curve centered at that frequency with bandwidth at least wider than the bars f Fig 12 19

20 Are the fllwing signals masked by the nise? 200Hz at 60dB 1KHz at 60dB Nise Masking Tne 20

21 Nise Masking Tne Are the fllwing signals masked by the nise? 200Hz at 60dB Yes! 1KHz at 60dB nise 21

22 Are the fllwing signals masked by the nise? 200Hz at 60dB N! 1KHz at 60dB Nise Masking Tne 22

23 Are the fllwing signals masked by the nise? 200Hz at 60dB 1KHz at 60dB N! Nise Masking Tne 23

24 Are the fllwing signals masked by the nise? 200Hz at 60dB 1KHz at 60dB N! Nise Masking Tne 24

25 Masking - Tempral Tempral Masking Masker effect persists fr tenths f a secnd Masker effect is acausal n ~ 2/100 timescales 25

26 Pitch JND [H. Fletcher. Rev. Md. Phys., 12(1):47 65, 1940]. JND = just nticeable difference change in stimulus that just elicits perceptual ntice where just means that a smaller variatins f stimulus cannt be discerned What can yu say abut the JND: Belw 1000 Hz? rughly cnstant ~ 3 Hz Abve 1000 Hz? rughly lg-lg linear Lg[Jnd(f 2 )] - Lg[ Jnd(f 1 )] ~ n (Lg[f 2 ] - Lg[f 1 ]) What is n? e.g. f 1 =2000 f 2 = = 10 4 ~ n( Lg 10 [2] ) ) n ~ 20 Suggests that as frequency increases brader frequency bands assigned t same length f cchlear tissue Remember cchlea mdel 26

27 JND experiment The fllwing audi files cntain a single tne playing fr 10 secnds. The sine starts at 200Hz, then changes t a higher frequency (201, 202, 203, 205, 210). This change ccurs after a number f nises : 1, 2, 3, 4, 5, 6, 7, 8 r 9. Can yu ntice when the change happens? 27

28 Critical Bands Decades f empirical study reveal that human audi frequency perceptin is quantized int < 30 critical bands f perceptually near-identical pitch classes crrespnding t ~equal length bands f cchlear tissue (neurns) 28

29 Critical Bands: Evidence [T. Painter and A. Spanias. Prc. IEEE, 88(4): , 2000.] Tne masking Nise (Fig. a & c) nise audibility threshld fr small bandwidth nise remains cnstant until tne frequency lcus falls away frm critical bandwidth Nise masking Tne (Fig. b & d) same effect with masker and maskee rles reversed 29

30 The Bark Scale [E. Zwicker. J. Acust. Sc.Am., 33(2):248, February 1961] Bark units: Unifrm JND scale fr frequency Maps frequency intervals int their respective critical band number 30

31 The Bark Scale Frequency-t-Bark functin First Principles vs. Empirical Mdeling [E. Zwicker. J. Acust. Sc.Am., 33(2):248, February 1961] B( f ) 13tan 1 ( f ) 3.5tan 1 (( f / 7500) 2 ) 31

32 Cmpressin pprtunities Cnsider the fllwing recrding Any ways t imprve the cmpressin? 32

33 Cmpressin pprtunities Zming in n a smaller prtin Any ways t imprve the cmpressin? 120 db Hz 200Hz 205Hz Masked Frequency 33

34 Cmpressin pprtunities Zming in n a smaller prtin Any ways t imprve the cmpressin? db JND: Culd nly represent integer frequency values Hz 200Hz 205Hz Frequency 34

35 Cmpressin pprtunities Zming in n a smaller prtin Any ways t imprve the cmpressin? 120 db Hz 200Hz 205Hz Frequency 35

36 Next Week Hw can we use what we knw abut human perceptin t cmpress music? Frequency hearing range Masking Tempral Spatial JND Barks 36

37 Big Ideas Sund is a pressure wave that makes the Cchlea vibrate with frequencies frm ~20Hz (at the tip) t ~20KHz (at the base) This vibratin is sinusidal (physics) This is why sund harmnics are best represented as sinusidal signals Masking Tempral A masker tne can mask anther tne that is present either right befre r a little after the masker Spatial A single tne can mask an entire frequency band (that cntains the tne) if its intensity is high enugh There are <30 such bands (Bark scale), and they are wider fr higher frequencies 37

38 Admin Lab 5 reprt due tmrrw On Thursday: Lab 6 Yu will be designing yur wn experiments T measure the range f frequencies yu can hear T perfrm spatial masking experiments 38

39 ESE250: Digital Audi Basics End Week 6 Lecture Human Psychacustics 39