Merimaa - Concer Hall Impulse Responses - Pori, Finland - Analisys Results (2005).pdf - Akustyka(1) - dzidzia2603

Concert Hall Impulse Responses — Pori, Finland:

Analysis results

Juha Merimaa 1 ,TimoPeltonen 2 , and Tapio Lokki 3

juha.merimaa@hut.fi, timo.peltonen@akukon.fi, tapio.lokki@hut.fi

1 Laboratory of Acoustics and Audio Signal Processing

Helsinki University of Technology

P.O.Box 3000, FI-02015 TKK, Finland

2 Akukon Oy Consulting Engineers

Kornetintie 4 A, FI-00380 Helsinki, Finland

3 Telecommunications Software and Multimedia Laboratory

Helsinki University of Technology

P.O.Box 5400, FI-02015 TKK, Finland

May 6, 2005

1 Introduction

This document presents analysis results of a published set of concert hall impulse responses. For

a detailed description of the hall, measurement positions, sound sources and microphones, as well

as the measurement procedure and post processing of the responses, see the reference document

(Merimaa et al., 2005). The responses and all documentation are available for download from

http://www.acoustics.hut.fi/projects/poririrs/ .

A major part of this document consists of individual analysis results for each pair of source-

receiver positions. The results include standard room acoustical parameters, spectrogram plots and

special directional analysis of the responses. The document is organized as follows. Section 2 de-

scribes computation and purpose of the listed room acoustical parameters and Section 3 introduces

the two types of ﬁgures used to illustrate the responses. The results for the individual responses

have been collected to the end of the document, each source-receiver combination being presented

on its own page.

2 Room acoustical parameters

The room acoustical parameters are a very traditional way of characterizing rooms or concert

halls. There is some controversy in the literature describing the relation of these parameters to the

actual perception of a listener. Nevertheless, the following subsections try to give some guidelines

for established interpretation of the results. In order to limit the analysis, we have only chosen to

present standardized parameters (ISO 3382, 1997) with the addition of Gade’s (1989a; 1989b; 1992)

support, describing the ability of a musician to hear him/herself when performing on the stage.

All parameters reported in this document have been calculated at octave bands. The ﬁltering

has been performed using ANSI S1.1-1986 standard ﬁlters as implemented in the Octave Matlab

toolbox (Couvreur, 1997). Except for the spatial parameters interaural cross-correlation and lateral

125 Hz 250 Hz 500 Hz 1 kHz 2 kHz 4 kHz 8 kHz

T 30 (s)

EDT (s)

(dB)

C 80 (dB)

−

IACC E

74 0

−

IACC L

89 0

94 0

LF P

37 0

34 0

LF SF

37 0

48 0

SNR (dB)

66 . 1

65 . 7

67 . 3

72 . 0 . 3 . 2 . 5

Table 1: Room acoustical parameters at octave bands averaged over all responses with receiver

positions in the audience area.

energy fraction, all other parameters are reported as an average of the values derived from both of

the DPA 4006 microphones.

All parameters except the signal-to-noise ratio (SNR, which is actually not a room acoustical

parameter but describes the measurements) have been calculated from the denoised responses (see

Merimaa et al., 2005, Section 5.1). This can be motivated as follows: Examination of the achieved

SNRs and start times of the denoising process reveals that the parameters that are evaluated over a

limited decay or period of time include in all cases only measured data. This leaves only strength and

clarity, where the computation of energy of the whole impulse response includes the extrapolated

parts. However, the denoising process extrapolates the responses in such a way, that the strength

and clarity over the denoised responses are actually better estimates of the real parameters than

using just the measured parts would yield.

The average parameters over all responses with receivers positions in the audience area are listed

in Table 1. This is the most appropriate way to characterize the hall itself. The parameters for the

individual source-receiver pairs presented in the end of this document should not be interpreted too

strictly. It has been shown that the parameters can vary considerably between individual locations

or even small displacements of measurement positions in a hall (Pelorson et al., 1992; Bradley, 1994;

Nielsen et al., 1998; Okano et al., 1998; de Vries et al., 2001), although it is common to assume

that the perception of a hall does not vary as much within the hall.

In the following subsections, each reported parameter and the applied computation methods

are described.

2.1 Reverberation time and early decay time

The reverberation time (RT) is the oldest and most common parameter describing concert hall

acoustics. It is deﬁned as the time that it takes for the sound inside a hall to decay 60 dB after a

source is turned oﬀ. Similarly, the early decay time (EDT) is deﬁned as the time during which the

ﬁrst 10 dB of the decay process occurs, multiplied by six.

In a perfectly diﬀuse hall, EDT and RT would always yield exactly the same values. In practice

they do, however, diﬀer to some degree, and EDT is more dependent on the geometry of a hall

and on the measurement position (Barron, 1995). Both RT and EDT aﬀect the subjective sense of

reverberance or liveness of a hall. EDT is considered a better descriptor for the running reverber-

ance, since during continuous sound most of the reverberation tail is masked by the sound itself.

Only when there is a longer break in the source signal, a listener will be able to hear the full decay

characterized by the RT.

According to Beranek (1996) the subjective reverberance is mainly determined by reverberation

time at mid and high frequencies above approximately 350 Hz. On the other hand, low frequency

reverberation creates a sense of warmth, which is more of a timbral attribute. Furthermore, exces-

sive low frequency reverberation can make a hall sound boomy. For describing the warmth, Beranek

(1996) has proposed bass ratio deﬁned as the ratio of average of RTs at the 125 and 250 Hz octave

bands to that at 500 and 1000 Hz. Gade (1989b) has proposed a similar measure computed from

EDTs at 250 and 500 Hz related to EDTs at 1 and 2 kHz.

The standard (ISO 3382, 1997) allows several ways to measure RT and EDT. We have chosen to

derive them from a least-squares line ﬁt to backward integrated squared impulse responses, which

gives an ensemble average of the decay curves that would be obtained with random noise samples

as an excitation (Schroeder, 1965). The RT and EDT are calculated from the slope of the ﬁtted

line. For determining the RT, the line was ﬁtted between

−

5and

−

35 dB points, which gives the

standardized

T 30

value (ISO 3382, 1997), and for EDT between 0 and

−

10 dB points relative to

the level of the direct sound.

2.2 Strength

is a parameter describing the amount of sound energy directed to a listening

position. It is deﬁned as the logarithmic ratio of the total energy of a measured response to that

produced by the same sound source with the same excitation at a distance of 10 m in a free

ﬁeld (ISO 3382, 1997). In addition to speciﬁc acoustical features of a hall, strength depends on

the distance from the sound source, as well as on the reverberation time (Barron and Lee, 1988).

Perceptually the strength (especially at mid-frequencies 500 and 1000 Hz, Beranek 1996) determines

the loudness of a hall. As a spectral parameter, the strength at diﬀerent frequency bands is, of

course, also related to the timbre.

Given the applied level calibration (Merimaa et al., 2005, Section 4.3), computation of the

strength is straightforward. The total energy at each octave band was divided by the energy of the

impulse response of the corresponding ﬁlter, and 10 dB was added to the resulting value.

2.3 Clarity

C 80 , the response is divided into the early and

late parts at 80 ms after the arrival of the direct sound. Clarity is highly correlated with the decay

parameters RT and EDT 1 and it also depends on the distance from the sound source somewhat

similar to

(Barron and Lee, 1988; Barron, 1995). The clarity is related to the perception of what

Beranek (1996) calls horizontal deﬁnition, deﬁned as “the degree to which sounds that follow one

another stand apart”.

The

C 80 values presented in this paper were calculated such that the responses were divided

into early and late part before the octave band ﬁltering. This way the time-domain spreading due

to the ﬁltering does not aﬀect the division.

2.4 Interaural cross-correlation and lateral energy fraction

Interaural cross-correlation (IACC) and lateral energy fraction (LF) are related to the spatial

properties of a hall. IACC can be calculated from impulse responses measured with a dummy

or a real head and it is deﬁned as the maximum of the normalized interaural cross-correlation

function over lags in the range of [ − 1 , 1] ms (ISO 3382, 1997). IACC is typically divided into

IACC E integrated over 0–80 ms from the arrival of the direct sound and to IACC L integrated over

80–1000 ms (Bradley, 1994; Hidaka et al., 1995). For determining the LF, an omnidirectional and

a ﬁgure-of-eight microphone are needed. LF is deﬁned as the fraction of the energy arriving from

lateral directions during the early part of a response (0–80 ms) (ISO 3382, 1997).

The IACC and LF are intended as measures of spatial impression. The spatial impression is

typically divided into auditory source width (ASW) and listener envelopment (LEV). ASW depends

1 The correlation is especially high between C 80 and the ratio of RT and EDT (Barron, 1995).

The strength factor

The clarity index is deﬁned as the ratio of early energy to the late (reverberant) energy expressed

in decibels (ISO 3382, 1997). In the clarity index

IACC E ]overthe

octave bands centered at 500, 1000, and 2000 Hz combined with strength at frequencies below 125

Hz as the best descriptors for ASW. However, despite the established nature of these parameters,

they do not always seem to be able to describe the perception, when acoustical environments of

considerably diﬀerent size are compared (Merimaa and Hess, 2004).

At low frequencies the [1 − IACC E ] values are always low since the relatively small distance

between the ears of the dummy head compared to the wavelength of sound results in a high

correlation. Nevertheless, at octave bands up to 1 kHz, the hall averages of [1

−

IACC E ]andLF

have been shown to correlate strongly (Bradley, 1994). With increasing frequency, IACC gains

more sensitivity to sound emanating from directions closer to the median plane than the (ideally)

frequency independent ﬁgure-of-eight directivity pattern used in LF. According to Okano et al.

(1998) IACC E describes the human spatial perception better than LF when these two parameters

disagree.

The IACC values reported in this paper were computed from the non-diﬀuse-ﬁeld-equalized

dummy head responses and they are reported in the form [1

−

IACC] such that high values describe

high diﬀuseness (low correlation).

The LF estimates were derived both from measurements with the Pearl TL-4 stereo back-to-back

cardioid microphone (LF P ) and the SoundField microphone system (LF SF ). The omnidirectional

energy was integrated over 0–80 ms and the lateral energy over 5–80 ms to exclude possible leakage

of the direct sound into the lateral signal (ISO 3382, 1997). With the Pearl microphone, the

responses of the left and right channels were subtracted from one another to obtain the ﬁgure-of-

eight directivity pattern and summed to obtain the omnidirectional reference response. In case

of the SoundField microphone, both directivity patterns were readily available, but the utilized

ﬁgure-of-eight channel Y was scaled by 1

2.5 Support

Support (ST) aims at describing a hall from a performing musician’s point of view. The ST

parameters were developed to describe the perceptual support, which is “the property which makes

the musician feel that he can hear himself and that it is not necessary to force the instrument to

develop the tone” (Gade, 1989a).

ST is deﬁned as the ratio of reﬂected energy to the emitted energy, as measured with an

omnidirectional microphone at a distance of 1 m from a sound source on the stage. The emitted

energy is integrated over the time interval of 0–10 ms after the arrival of the direct sound, including

typically the direct sound and the ﬁrst ﬂoor reﬂection. For computing the reﬂected energy, Gade

(1989a,b) proposed originally two time intervals: 20–100 ms (ST1) and 20–200 ms (ST2). In a

later publication (Gade, 1992), ST1 was renamed ST early and ST late including reﬂected energy

integrated over 100–1000 ms was introduced. Furthermore, ST2 was replaced by ST total ,whichcan

be calculated as the (linear) sum of ST early and ST late .

The ST early and ST late values for the three source positions are listed in Table 2. These param-

eters were calculated from responses which were excluded from the public database due to little use

for anything apart from the support calculation. The responses were measured with the DPA 4006

microphone pair at a distance of 1 m from each source position and the parameters computed from

2 Bradley and Soulodre (1995b) have also proposed late lateral energy (as opposed to energy fraction) for measuring

envelopment.

mainly on the early reﬂections characterized by IACC E and LF, wheras the envelopment is created

by diﬀuse late reverberation as measured with IACC L (Morimoto and Maekawa, 1989; Bradley

and Soulodre, 1995a) 2 . Okano et al. (1998) have proposed the average of [1

/ √ 2 to compensate for the B-format gain convention. It

is interesting to notice that the results diﬀer considerably. Part of these diﬀerences may be due

to nonideal directivity patterns of the microphones. Furthermore, measurements of a SoundField

MKV system (Farina, 2001) have shown another source of error caused by frequency dependent

variations in the relative gains of the omnidirectional and ﬁgure-of-eight channels. This result leads

the authors to believe that the LF P values are more reliable.

ST early

250 Hz 500 Hz 1 kHz

2 kHz

4 kHz

−

Avg.

−

ST late

250 Hz 500 Hz 1 kHz

2 kHz

4 kHz

− 11 . 2

− 12 . 1

− 12 . 0

− 10 . 4

− 10 . 6

−

Avg

− 12 . 8

− 12 . 3

− 11 . 9

− 11 . 1

− 10 . 8

Table 2: Support (in dB) at each source position.

both microphones were averaged. Only the omnidirectional sound source (as opposed to having

also a subwoofer) was used and for this reason the ST at the 125 Hz octave band has been omitted

from the results. Furthermore, the source cannot be considered omnidirectional at the omitted 8

kHz octave band and the listed 4 kHz results should also be interpreted with some care while the

results are prone to random errors depending on the direction of the measurement position relative

to the sound source.

2.6 Signal-to-noise ratio

As mentioned earlier, signal-to-noise ratio (SNR) is actually not a room acoustical parameter but

describes the measurements. The listed SNR values were calculated as the ratio of the peak value

of a response to the background noise level averaged over 10 % of the end of a response prior to

denoising. The DPA 4006 responses were used, and in each case it was veriﬁed that the 10 % of

the samples were indeed background noise.

3Fgus

The ﬁgures presented for each source-receiver pair in the end of this document consist of spectro-

grams and a special directional analysis of the early responses.

3.1 Spectrograms

The spectrograms provide a time-frequency representation of the decay of a response. They were

calculated from the rightmost (reference) DPA 4006 microphone measurements after applying the

denoising procedure (Merimaa et al., 2005, Section 5.1). The computation was performed using

1024 sample FFT with 50 % overlapping Hann windowed time frames. The resolution is thus the

same as that used in the denoising. The energy of each time-frequency component is plotted on a

dB scale. The frequency range of the plots has been limited to 20 kHz and the spectrograms are

presented over the full length of the waveform ﬁles in the database. Furthermore, the frequency-

dependent starting point of the extrapolated exponentially decaying random noise created in the

denoising process is shown with a solid black line on top of the spectrograms.

3.2 Directional analysis

The directional analysis plots illustrate the direction dependent arrival of sound to a measurement

position during a time period of 100 ms starting from slightly before the arrival of the direct sound.

Merimaa - Concer Hall Impulse Responses - Pori, Finland - Analisys Results (2005).pdf

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: