Calibration of Consonant Perception in Room Reverberation

K. Ueno (Institute of Industrial Science, Univ. of Tokyo)

N. Kopčo and B. G. Shinn-Cunningham (Hearing Research Center, Boston Univ.)

Introduction

Our auditory process is usually assumed to be static and fixed, dependent only on the input signals rather than on the state of the listener.

We naturally and fluidly compensate for

many interfering effects in everyday environments.

How do listeners calibrate auditory perception to acoustic interference?

Outline of the Study

PURPOSE: To explore how listeners calibrate auditory perception to room reverberation.

STRATEGY: Measure the effect of sudden changes of reverberation on speech perception.

Carrier phrase (RevC) Target (RevT)

Target (RevT)Carrier phrase (RevC) --- Lower performance

Un-matching reverberation

--- Higher performance

Matching reverberation

HYPOTHESIS: Consonants identification performance should be better when listeners have consistent room experience just prior to a test sound.

Stimuli

VC1 VC2 - - - - VC

　 * Rev-C ----- *Rev-T

Carrier phraseCarrier phrase TargetTarget

Speech source: VC (Vowel-Consonant) syllables with 16 consonants preceded by ‘o’ (/a/) ok, ot, op, of, od, og, ob, ov, oth(v), om, on, ong, oz, oth(uv), os, osh Two male and One female Recordings from corpus and a past study

Binaural room IR (BRIR): R1, R2, Anechoic

Test sound: VC*BRIR

Binaural room impulse responses R1: at relatively closer point

(12m) to the sound source in very reverberant church.

… reverberant

R2: at second balcony in a large concert hall (33m) … reverberant

Pseudo-anechoic BRIR are processed from R1 BRIR by a 5-ms time window. … dry (clear)

R1,Lch

R2,Lch

R1R2

Binaural room impulse responses R1: at relatively closer point

(12m) to the sound source in very reverberant church.

… reverberant

R2: at second balcony in a large concert hall (33m) … reverberant

Pseudo-anechoic BRIR are processed from R1 BRIR by a 5-ms time window. … dry (clear)

R1,Lch

R2,Lch

Processed for Pseudo-Anechoic HRTF

Experimental Design and Procedure

Test signals were presented with insert headphones. Subject’s responses for the final VCs were obtained by GUI using

16 graphical buttons labeled with the VCs. Number of VCs (2 or 4) in the carrier was fixed throughout blocks

of trials. Stimuli set (10 VCs x 3 talkers x 3 conditions = 90 trials in total)

were randomly presented in each block, repeated twice for each subject.

Subjects: 14 Native English speakers Percent-correct target identification scores were calculated for

each condition and subject.

tt =0.8 s

2 VCs carrier ----4 VCs carrier ----

VC1 VC2 VCVC1 VC2 VC3 VC4 VC

Rev-C ----- Rev-T

Carrier phraseCarrier phrase TargetTarget

t t t tRev-T

R1 R2 AE

Rev-C2VCs or 4VCs

AE RAE RAE -R1 Rm Rnm -R2 Rnm Rm -

Experimental Results

Carrier Reverberation

2VCs

% C

orre

ct ta

rget

iden

tific

atio

nRev-T

R1 R2 AE

Rev-C2VCs or 4VCs

AE RAE RAE -R1 Rm Rnm -A2 Rnm Rm -

4VCs

RmRm RnmRnm RAERAE

****

***

○： Rev-T ＝ R1 ，●： Rev-T=R2

The effect of Rev-C is significant only with Rev-T=R2 (p<.0001): performance with matching reverberation is significantly higher than unmatching rev. with Rev-T=R2.

The effect of the carrier length is not significant.

Condition means of the PC: across 14 subjects and two repetitions

Error bars: showing 95 % confidence intervals for mean within subject (14 data)

Analysis of BNIR - reverberation

Frequency [Hz]Reverberation Energy

(Rev(50ms-)/Dir(0-50ms))

Frequency [Hz]

Reverberation Time (T60)

R1 R2

Frequency [Hz] Frequency [Hz]

FFT of early 100ms

Relative level [dB]

SNR and STI

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=125 Hz

MT

F

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=250 Hz

MT

F

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=500 Hz

MT

F

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=1k Hz

MT

F

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=2k Hz

MT

F

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=4k Hz

MT

F

1.0 2.0 4.0 8.00

0.2

0.4

0.6

0.8

1Fc=8k Hz

Modulation frequency [Hz]

MT

F

125 250 500 1k 2k 4k 8k-5

0

5

10

15

Frequency (Fc) [Hz]

SN

R [

dB]

AE (STI=0.99)R1 (STI=0.45)R2 (STI=0.45)

R1 R2

Summary

Calibration to room reverberation improved consonant perception in one (but not in the other) room explored in this study.

The two rooms differ in several acoustic characteristics, which might be the cause of this effect.

The calibration occurs quickly, after just a few words.

Thank you for your attention!