Database: Recordings of a Loudspeaker Orchestra with Multi-Channel Microphone Arrays for the Evaluation of Spatial Audio Method [original]

Database: Recordings of a

Loudspeaker Orchestra with

Multi-Channel Microphone Arrays

for the Evaluation of Spatial Audio

Methods

—

David Ackermann, Julian Domann, Fabian Brinkmann,

Johannes M. Arend, Stefan Weinzierl

Audio Communication Group, TU Berlin

Einsteinufer 17c, 10587 Berlin, Germany

david.ackermann@tu-berlin.de

stefan.weinzierl@tu-berlin.de

December 13, 2022

Figure 1: Set-up of the 18-channel loudspeaker orchestra on the stage in the large hall

of the Berliner Philharmonie, with the slightly elevated listener position in the

first row of the audience area (highlighted by the bright rectangle).

General Information

This document describes the recordings of a loudspeaker orchestra with multi-channel

microphone arrays for the evaluation of spatial audio methods. The database is provided

under a Creative Commons BY-SA licence, giving you the freedom to redistribute and

edit it for non-commercial purposes. For more information visit creativecommons.org/

licenses/by-sa/4.0.

The documentation is focused on the structure and content of the database. More

information on the acquisition, processing and validation can be found in the accompa-

nying publication in the Journal of the Audio Engineering Society (JAES). If you use

this database, please cite:

D. Ackermann, J. Domann, F. Brinkmann, J. M. Arend, M. Schneider, C. P¨orschmann,

and S. Weinzierl (2023): ”Recordings of a loudspeaker orchestra with multi-channel mi-

crophone arrays for the evaluation of spatial audio methods” Journal of the Audio En-

gineering Society, DOI: doi.org/10.17743/jaes.2022.0059

D. Ackermann, J. Domann, F. Brinkmann, J. M. Arend, and S. Weinzierl (2022):

”Recordings of a Loudspeaker Orchestra with Multi-Channel Microphone Arrays for the

Evaluation of Spatial Audio Methods”, 2022, DOI: dx.doi.org/10.14279/depositonce-

15774

Database description

Database

0 Documentation.pdf

1 Recordings BOL.zip

1 Recordings BPH.zip

2 IR BOL.zip

2 IR BPH.zip

3 AdditionalData.zip

Figure 2: Data structure of the database.

The presented data set includes the record-

ings of an 18-channel loudspeaker orchestra

in two different rooms. As source material,

an arrangement for 18 individual orchestral

instruments of the sixth movement of the

suite Children’s Corner, L. 113, by Claude

Debussy, recorded in an anechoic chamber,

was used. The loudspeaker orchestra was

recorded with 8 different multichannel mi-

crophones placed at exactly the same pos-

tion using a laser device. The database in-

cludes an omnidirectional measurement mi-

crophone, head-and-torso simulators (HATS), equatorial microphone arrays, as well as

first-order ambisonics (FOA) and higher-order ambisonics (HOA) microphones (cf. Ta-

ble 2). In addition, impulse responses were measured for each sound source and micro-

phone. In the case of the FABIAN head-and-torso simulator (Listener1), BRIRs for head

rotations to the left and right were measured in the range of ±45◦and with a resolution

of 1◦. Furthermore, 3D room models, an excerpt of the used anechoic audio recordings

and pictures of the measurement setup are provided.

The general data structure is outlined in Figure 2. The database consists of 4 compo-

nents: (0) this documentation, (1) recordings of the loudspeaker orchestra as SOFA files,

(2) IRs of the loudspeaker orchestra as SOFA files and (3) additional data of the rooms.

For each room and microphone, the database contains two files: One for the impulse

responses (IRs) and one for the loudspeaker orchestra recording. Note that during the

loudspeaker orchestra recording, all emitters played the anechoic audio simultaneously.

All data were measured with a sampling frequency of fS= 48 kHz.

Terminology

From now on, the microphones are referred to as Listeners, and each capsule of a mi-

crophone is considered a Receiver. The loudspeaker orchestra is termed the Source with

each loudspeaker being an Emitter. This follows the naming convention of the SOFA

standard, and meta data such as the position of an emitter or orientation of a listener

are stored in correspondingly named fields in the corresponding SOFA files. For an in-

troduction of the SOFA format, the reader is referred to Majdak et al. [1]. The standard

is published as AES69-2022 [2].

1 Recordings

The recordings are available for each room and listener combination in separate SOFA

files within the SingleRoomSRIR convention. The SOFA files are stored in the archives

’1 Recordings BPH.zip’ (BPH - Berliner Philharmonie) and ’1 Recordings BOL.zip’

(BOL - Berlin Open Lab). The archives contain the files:

1_BOL_Listener1_recordings.sofa

...

1_BPH_Listener9_recordings.sofa

They can be loaded for example using the Matlab/Octave API [3]:

obj = SOFAload(’1_BOL_Listener1_recordings.sofa’);

or with the sofar Python package [4]:

sofa = sofar.read_sofa(’1_BOL_Listener1_recordings.sofa’)

After loading, the time data of the recording can be accessed with the field directory

obj.Data.IR or sofa.Data IR. The field has the dimension MxRxN, where M (measure-

ment) is always 1, R (receiver) corresponds to the number of capsules of the associated

microphone array and N indicates the length of the recording in samples. Since the ane-

choic audio signals were played back at the same time from the 18 loudspeakers, the 18

emitters are considered as one source in this convention. However, the relative position

of the emitters to the source position, which was defined as the center of the loudspeaker

orchestra, can be retrieved from the related metadata EmitterPosition.

2 IR

The IRs are stored for each room and listener combination separately in dedicated SOFA

files within the SingleRoomMIMOSRIR convention. The SOFA files are stored in the

archives ’1 IR BPH.zip’ (BPH - Berliner Philharmonie) and ’1 IR BOL.zip’ (BOL -

Berlin Open Lab). The archives contain the files:

2_BOL_Listener1_RIR.sofa

...

2_BPH_Listener9_RIR.sofa

They can be loaded for example using the Matlab/Octave API [3]:

obj = SOFAload(’2_BOL_Listener1_RIR.sofa’);

or with the sofar Python package [4]:

sofa = sofar.read_sofa(’2_BOL_Listener1_RIR.sofa’)

The IRs are stored in the field obj.Data.IR and sofa.Data IR with the dimension

MxRxNxE, where M is 1 in most cases, except for the listener L1 (FABIAN), where M

denotes the head-above-torso orientation between -45◦to 45◦in 1◦steps, which corre-

sponds to M=91 measurements. Furthermore, the dimension R (receiver) corresponds

to the number of capsules of the microphone array or dummy head, N to the length

of the RIRs, and E (emitter) to the number of loudspeakers, which is always 18. The

musical instruments assigned to the emitters can be determined from the metadata of

the file. Furthermore, the metadata contain additional information about the room and

the measurement conditions.

3 AdditionalData

3 BOL

BOL Stimuli

BOL E01.wav

...

BOL Pictures

BOL.skp

3 BPH

BPH Stimuli

BPH E01.wav

...

BPH Pictures

BPH.skp

3 OtherData

LoudspeakerReferencePoints

DirectivityIndex

Figure 3: Data structure of the additional

data.

All the additional data can be downloaded

as one ZIP file. The data structure is

shown in Figure 3. An excerpt of the

anechoic recordings of the sixth move-

ment of the suite Children’s Corner, L.

113, by Claude Debussy is provided in

the folder RoomName_Stimuli. The nam-

ing of the audio data follows the scheme

RoomName_EmitterNr.wav, the detailed as-

signment to the musical instruments can

be found in Table 1. For each room, the

balance and timbre of the loudspeaker or-

chestra was adjusted to the room acoustic

properties by a professional sound engineer.

Thus, one anechoic recording is provided

per room, each containing the respective

level and timbre settings. For comparing

IR based auralizations against recordings,

it is thus sufficient to convolve the anechoic

recordings with the IRs without any addi-

tional filtering.

For both rooms, a 3D model of the scene

saved as SketchUp file is provided. Each

file contains the names, positions, and orientations of the sources and receivers.

In addition, pictures are provided showing the environment, the loudspeaker orchestra

and the various microphones. These data are stored separately for both spaces, the

Berliner Philharmonie (BPH) and the Berlin Open Lab (BOL).

Furthermore, images are included showing the reference point for determining the posi-

tion of all speakers with a red laser dot and the directivity index over the frequency for

the octave bands from 200 Hz to 8000 Hz of the musical instruments is stored.

Table 1: Assignments of the loudspeakers to the instruments of the audio material

Emitter Instrument Loudspeaker

E01 Violin 1 Neumann KH 120

E02 Violin 2 Neumann KH 120

E03 Viola Neumann KH 120

E04 Cello Neumann KH 120

E05 Double bass Genelec 1031

E06 Picolo flute Genelec 8020

E07 Flute Genelec 8020

E08 Oboe Genelec 8020

E09 Clarinet K+H O300D

E10 Bass clarinet K+H O300D

E11 Bassoon Neumann KH 120

E12 French horn Genelec 1031

E13 Trumpet Genelec 8331

E14 Snare drum HEDD Type 07

E15 Bass drum Genelec 8020 + KRK 12s

E16 Triangle Genelec 8020

E17 Crash cymbals HEDD Type 07

E18 Harp Genelec 8020

Loading more pages...