2015 23rd European Signal Processing Conference (EUSIPCO)

A METHODOLOGY FOR ESTIMATING THE RADIATION PATTERN OF A VIOLIN

DURING THE PERFORMANCE

Antonio Canclini, Luca Mucci, Fabio Antonacci, Augusto Sarti, and Stefano Tubaro

Politecnico di Milano, Dipartimento di Elettronica, Informazione e Bioingegneria, Milano, Italy

ABSTRACT other systems measure the directivity pattern of an instrument

We propose a method for the estimation of the three-dimensional played by a violinist. In [5] a large semicircular microphone
radiation pattern of violins, during the performance of a mu- array is used to measure the sound radiated by a violin, while
sician. A microphone array captures the energy radiated by the player sits on a rotating stool. In [6] the musician stands
the violin in different directions using beamforming based on at the center of an anechoic room, holding the violin in a pre-
sub-arrays. The 3D radiation pattern is estimated allowing the scribed position, and the acoustic energy is measured at 22
musician to freely move. In particular, a tracking system esti- points surrounding it. A similar approach is followed in [7],
mates the position and orientation of the violin. The adopted where a scanning probe microphone is used to measure the
system can be also used in a mildly reverberant environment, soundfield over planar regions. In [8] the violin is also kept
thus allowing the musician to play in a natural fashion. The still on a rigid structure. Unfortunately, this solution prevents
experimental results prove the accuracy and the effectiveness the musician to play in a natural fashion.
of the method. All the mentioned techniques rely on measurements con-
ducted in anechoic rooms, which ensure to capture the sole
Index Terms— Musical acoustics, plenacoustic analysis, direct sound radiation. This scenario, while ideal in case of
radiation pattern artificial excitation, presents some issues for the violinist. In-
deed, it exposes the player only to the dry violin sound, which
1. INTRODUCTION typically provokes an annoying sensation that may impact on
the naturalness of the performance. In this paper we propose
Several acoustic properties of the violin are studied. A par- a novel technique for measuring the 3D radiation pattern of a
ticularly interesting one is the directivity or radiation pattern, violin. A compact rectangular microphone array (plenacous-
which gives the angular dependency of the sound energy ra- tic camera) is used to sense the directional components of the
diated from the instrument in the far field. The knowledge of wave field. This is accomplished by the plenacoustic analysis
the directivity allows us to infer how the sound will interact of the wavefield proposed in [9], useful to estimate the direc-
with the environment. The radiation pattern of the violin is tional acoustic energy at multiple points in space. This anal-
a complex function that depends on several factors (e.g., ma- ysis is based on beamforming, and enables the measurement
terials and shapes), and it is difficult to predict, as shown in of the radiation pattern even in mildly reverberant rooms. A
[1], and subtleties related to it can arise complex phenomena simple tracking system, composed of a depth map camera and
such as the “directional tone color” [2], which cause different a gyroscope, is used to track the position and the orientation
notes to be perceived as coming from different directions. of the musician with respect to the microphones. Each pose
In this paper we propose a methodology for measuring the assumed by the violinist contributes to the portion of radiation
radiation pattern during the performance. Several techniques pattern corresponding to the part of the instrument “exposed”
have been proposed in the literature. Most of them are based to the plenacoustic camera. A reference microphone mounted
on artificial excitation mechanisms: in [2] loudspeakers are in proximity of the violin is used to normalize the energy.
adopted for exciting the violin; a force hammer is used in [3];
while the method in [4] relies on a mechanical bowing ma- 2. NOTATION AND PROBLEM FORMULATION
chine. Despite being characterized by a controlled and repeat-
able excitation of the instrument, these methods do not con- 2.1. Violin coordinate system
sider the effect of human body on sound radiation. In order to
account for the modifications introduced by the human body, As we are interested in estimating the far field radiation, we
consider the violin as a point source located in the center of
THIS RESEARCH ACTIVITY HAS BEEN PARTIALLY FUNDED
BY THE CULTURAL DISTRICT OF THE PROVINCE OF CREMONA,
mass of the instrument. This position approximately corre-
ITALY, A FONDAZIONE CARIPLO PROJECT, AND BY THE ARVEDI- sponds to the bridge location, and it is considered as the ori-
BUSCHINI FOUNDATION. gin of the violin coordinate system. The coordinate axes are

978-0-9928626-3-3/15/$31.00 ©2015
Authorized licensed use limited Petrobras. Downloaded on 1546
to: IEEE July 16,2020 at 13:23:22 UTC from IEEE Xplore. Restrictions apply.
 2015 23rd European Signal Processing Conference (EUSIPCO)

oriented as in Fig. 1, with the bridge lying on the xz plane. In

the paper, we will work also with spherical coordinates, i.e. a
point x is defined by azimuth φ, elevation θ, and distance r
from the origin of the reference frame.

Fig. 2: Plenacoustic analysis.

Fig. 1: Violin coordinate system.

in the global reference system1 . Let s be the position of the
violin in the global coordinate system, and R be the 3×3 ma-
trix defining its rotation with respect to the global reference
2.2. Radiation pattern frame. The point xi = [xi , yi , zi ]T in the violin coordinate
system is
The radiation pattern is defined starting from the far field so-
lution of the Rayleigh’s first integral [10]. Under the far field xi = [xi , y i , z i ]T = R−1 (xi − s) . (3)
assumption, the temporal Fourier transform of the radiated
sound field at x is given by The spherical coordinates of xi are φi , θi , ri and can be eas-
ily computed from xi . Under the assumption that only the
P (x, ω) = g(r, ω)S(ω)D(φ, θ, ω) , (1) violin is present on the acoustic scene, we have that the ple-
nacoustic function at xi is given by
where S(ω) is the temporal Fourier transform of the violin
source signal, and ω the angular frequency; the Green’s func- L(xi , ui , ω) = P (xi , ω) = g(ri , ω)S(ω)D(φi , θi , ω) , (4)
tion
ω
ej c r where ui is the direction of the ray originating at s and pass-
g(r, ω) = (2) ing through xi . The radiation pattern is obtained from (4) as
r
 
accounts for the propagation delay and attenuation and c is    
D(φi , θi , ω) =  L(xi , ui , ω) 
the speed of sound. The term D(φ, θ, ω) is the directivity  g(ri , ω)S(ω) 
function. The radiation pattern is |D(φ, θ, ω)|, and describes (5)
|L(xi , ui , ω)|
the intensity of the sound field emitted towards direction = xi − s .
(φ, θ) at frequency ω. For an omnidirectional point source |S(ω)|
|D(φ, θ, ω)| = 1/(4π).
3. DIRECTIVITY ESTIMATION
2.3. Plenacoustic analysis In this
 Section we  describe the proposed method for obtain-
The goal of this work is to estimate |D(φ, θ, ω)| at all the fre- ing D(φi , θi , ω) from the integration of data retrieved from
quencies of interest, over a regular grid of angles covering the the plenacoustic camera; the reference microphone; and the
full spherical domain. To do so, we rely on the plenacoustic tracking system. All the measurement devices are mutually
analysis [9], a powerful tool for determining the directional synchronized, in order to obtain consistent data.
components of the sound field at multiple points in space.
In particular, this analysis estimates the plane-wave compo- 3.1. Reference microphone
nents L(x, u, ω) ∈ C propagating towards direction u and
contributing to the sound-field in x. This goal is achieved by A reference signal s(t) is captured by a microphone located
means of space-time processing of data acquired from the ple- close to the bridge of the violin. As the violin moves and the
nacoustic camera. Notice that during the performance the vi- sound changes during the musical performance, we operate
olin moves. It is therefore needed to adopt a global reference a short-time analysis on s(t). We assume the violin as static
frame, which is integral with the camera, and for convenience and s(t) as stationary during the K samples long observa-
is centered on the bottom-left corner of the camera and ori- tion window. The extracted frame is sv (n), where n is the
ented as in Fig. 2. We estimate L(x, u, ω) ∈ C on a regular 1 Primitives in the violin and global reference frames are identified by ·

rectangular grid defined by points xi , i = 1, . . . N , expressed and · symbols, respectively

Authorized licensed use limited to: Petrobras. Downloaded on 1547

July 16,2020 at 13:23:22 UTC from IEEE Xplore. Restrictions apply.
 2015 23rd European Signal Processing Conference (EUSIPCO)

discrete-time index and v is the frame index. In order to dis- estimated through delay-and-sum beamforming [13]. More
card frames not associated to a violin sound, or with a poor specifically, we estimate the energy radiated along direction
signal-to-noise-ratio, we perform a frame selection. At this x i − sv
purpose, for each frame we compute the short-time energy ui = (7)
xi − sv 

 K−1 through the pseudo-spectrum ψv (xi , ui , ω), computed as in
1 
ev =  sv (n)2 (6) [14]. The magnitude of the plenacoustic function is given by
K n=0 
|Lv (xi , ui , ω)| = ψv (xi , ui , ω). (8)
and the harmonic ratio HRv , computed using Eq.(2.33) in
[11]. The value HRv = 1 is reached for purely harmonic 3.4. Data integration
frames. The vth frame is selected if and only if ev ≥ Te and
HRv ≥ THR , where Te and THR are prescribed acceptance Once we have processed the frame v, we obtain a set of sam-
thresholds. For acceptable frames, we compute the energy of ples of the radiance pattern |D(φ, θ, ω)|. More specifically,
the signal at the frequency bins ωk , to obtain Sv (ωk ). we obtain the following samples:
 
Dv (φi , θi , ωk ) = xi − sv  |Lv (xi , ui , ωk )| , (9)
3.2. Tracking system |Sv (ωk )|
for i = 1, . . . N . In order to be robust against measurement
In this paragraph we describe the two components of the
noise, we consider only the frequency bins ωk corresponding
tracking system, namely an Inertial Measurement Unit (IMU)
to the harmonic peaks in the spectrum whose energy is not be-
and a depth map camera. Both the devices are synchronized
low 20 dB that of the main peak. We conclude the measuring
with the audio recording device. Therefore, all the measure-
session when the violinist moved and rotated in such a way
ments can be related to the current audio frame v.
that the radiance pattern has been estimated on the whole 3D
3.2.1. IMU angular domain. The gathered data is finally averaged and in-
terpolated to obtain the full 3D pattern, sampled on a regular
A 9-degrees-of-freedom IMU device is used to determine the
grid of Nφ × Nθ = 72 × 36 angular positions.
orientation of the instrument. The device is attached to the
chin-rest and positioned such that its local reference frame is
oriented as in Fig. 1. The orientation of the violin at frame 4. RESULTS
v is retrieved through the Euler angles, i.e. yaw αv , pitch βv ,
In order to assess the accuracy of the estimation, we mea-
and roll γv , as in [12]. From the Euler angles it is possible to
sured the radiation pattern of two different types of acoustic
retrieve the rotation matrix Rv at frame v.
sources. First the pattern of a commercial loudspeaker has
3.2.2. Depth map camera been measured in order to validate the methodology. Then
we estimated the radiation pattern of a standard full-size stu-
In order to track the violin position with respect to the global
dent violin.
coordinate system, we employ a Microsoft Kinect. The
Kinect produces a depth map that is used to fit the skeleton
model of the violin player. We estimate the back edge of 4.1. Acquisition setup
the chin rest as the middle point between the head and neck Experiments were conducted in a low-reverberation chamber,
joints of the skeleton model. The bridge location is then esti- with reverberation time T60 ≈ 50 ms. We used a T-Bone
mated as the middle point along a segment starting from the Ovid CC 100 super-cardiod condenser transducer to measure
chin-rest, of length 34 cm and oriented as the y axis of the |S(ω)|. The plenacoustic camera has been realized using 32
violin. Beyerdynamics MM1 measurement microphones, character-
ized by a flat frequency response and by a omnidirectional
3.3. Plenacoustic camera polar pattern. The microphones are arranged on a 4 × 8 grid,
thus N = 2 × 6 = 12 sub-arrays are available. The mi-
The ideal plenacoustic camera is implemented by Nr × Nc = crophone spacing is 7 cm in both the horizontal and vertical
Nm omnidirectional microphones placed on a regular grid, directions, corresponding to a spatial aliasing frequency of
all synchronized with the reference one located on the violin. about 2.5 kHz. The Kinect was placed in a calibrated posi-
We group the microphones in subarrays of M = 3×3 sensors tion below the plenacoustic camera. The recording sampling
each. Sub-arrays are maximally overlapped, i.e. N = (Nr − frequency was set to 48 kHz. Audio frames were extracted
2) × (Nc − 2) sub-arrays are available. Let xi be the center using a Hann window of K = 8192 samples, with 50% over-
of the ith subarray, and mij , j = 1, . . . M , the positions of lap. The acceptance thresholds defined in Sec. 3.1 were set to
the sensors within the subarray. The plenacoustic function is THR = 0.8 and Te = 0.25.

Authorized licensed use limited to: Petrobras. Downloaded on 1548

July 16,2020 at 13:23:22 UTC from IEEE Xplore. Restrictions apply.
 2015 23rd European Signal Processing Conference (EUSIPCO)

0° 0°
scenario we tested the whole system. We left the musician
free to move, assuming the positions needed for covering the
° °
30 330 30° 330°

-10dB -10dB
60° 300° 60° 300° positions of interest, holding the instrument in a standard and
-20dB -20dB
fixed location with respect to his head. We asked him to keep
90° 270° 90° 270°
each position for about 2 s, while playing the open D string.
120
°
240
°
120° 240°
To avoid unnatural poses, we restricted the analysis to the top
° ° ° °
half sphere of the radiation pattern (z ≥ 0), corresponding to
150 210 150 210
180
°
180° the front side of the violin.
(a) 250 Hz (b) 630 Hz Results are reported in Fig. 4, where the first row refers to
0 °
0°
the horizontal plane and the last one shows the full 3D pattern
30° 330° 30° 330°
as a function of azimuth φ and elevation θ. We show the radi-
-10dB -10dB
60°
-20dB
300° 60°
-20dB
300° ation diagrams of the 5 highest harmonic components of the
D note, from 294 Hz (the fundamental frequency) to 2058 Hz
90° 270° 90° 270°
(seventh harm.). Diagrams are normalized at each frequency
120° 240° 120° 240° with respect to the maximum energy radiated in all the sensed
150° 210° 150° 210°
directions. The dashed line in the first row of Fig. 4 is the
180° 180°
pattern obtained in the controlled scenario described before.
(c) 1250 Hz (d) 2000 Hz The very good matching between the measured pattern and
the reference one reveals the accuracy of the tracking system.
Fig. 3: Genelec 1029A: measured pattern (continuous line) vs
We observe that, at 294 Hz and 588 Hz, the radiation
reference pattern (dashed line).
pattern is mostly omnidirectional, except for a slight energy
damping in correspondence of the violinist’s head and neck.
4.2. Radiation of a commercial loudspeaker As the frequency increases, the pattern becomes more direc-
tive and exhibits more irregular shapes. This behavior agrees
We measured the horizontal radiation pattern of a Genelec with the predictions and results provided by Weinreich in
1029A speaker. Its location is known and it is at a distance [2], where the transition from an isotropic to an anisotropic
of 1.5 m from the plenacoustic camera. The speaker is po- radiation is found to occur approximately at 800 Hz. Indeed,
sitioned on a Outline ET250-3D electronic turntable. The at 1176 Hz there is a clear preferred direction of emission,
acoustic center of the loudspeaker (i.e., the midpoint between while at 2058 Hz the pattern presents three main radiation
the woofer and the tweeter) was at the same height of the cam- lobes.
era center. The speaker moved on 36 orientations on the full Results similar to Fig. 4 are shown in Figs. 4.17-19 of
circle. For each location, it emitted 2 s of white noise. Since [15]. For instance, the energy damping at 588 Hz in the range
orientation of the speaker and energy of the emitted signal are 30◦ − 60◦ , in the horizontal plane, is similar to that in Fig.
known, we bypassed the frame selection and tracking stages. 4.18 of [15] at 550, Hz. Moreover the main direction of emis-
Comparing the measured pattern with the ground-truth pro- sion at 1176 Hz is around 300◦ in the horizontal plane, which
vided by the manufacturer, the resulting average error is be- corresponds to that reported in [15]. A similar behavior is
low 3 dB in the frequency range 100 Hz - 2.5 kHz. Examples also described in [6]. A direct comparison is not possible,
are reported in Fig. 3 at four frequency values. as the results in [15] and [6] are relative to different violins.
Finally, it is worth noticing that the effect of the body of the
musician is visible at all frequencies. However, its impact on
4.3. Radiation of the violin the radiation becomes relevant from 882 Hz. This frequency
We estimated the radiation pattern of a full-size student violin corresponds to a wavelength of 39 cm, which starts to be com-
in two different scenarios. First, we instructed the musician parable with the head size.
to play holding the instrument in predefined positions. Since
pose and rotation of the violin were known, we neglected the 5. CONCLUSIONS
data coming from the tracking system. Due to the intrinsic
limitations in accurately positioning the instrument held by a We have presented a novel methodology to measure the radi-
person, we limited this analysis to the plane xy. The musi- ation pattern of a violin during the performance, by means of
cian stood about 1 m away from the camera, holding the vio- plenacoustic analysis of the radiated soundfield. The radiation
lin with the plate parallel to the floor. We asked him to rotate pattern is estimated incrementally as the violinist moves and
in 18 angular prescribed positions, pivoting on the bridge. In rotates. The directional energy of the soundfield is measured
each position, the violinist played the open D string for ap- by means of a plenacoustic camera, while the violin position
proximately 2 s. This way, we obtained a reference measure and rotation is tracked through a Kinect and a IMU device.
of the radiation pattern on the horizontal plane. In the second The proposed method presents the advantage of allowing the

Authorized licensed use limited to: Petrobras. Downloaded on 1549

July 16,2020 at 13:23:22 UTC from IEEE Xplore. Restrictions apply.
 2015 23rd European Signal Processing Conference (EUSIPCO)

90° 90° 90° 90° 90°

120° 60° 120° 60° 120° 60° 120° 60° 120° 60°

150° 30° 150° 30° 150° 30° 150° 30° 150° 30°
-10dB -10dB -10dB -10dB -10dB
--20dB --20dB --20dB --20dB --20dB

180° ° °
0180 ° °
0180 ° °
0180 ° °
0180 0°

210° 330° 210° 330° 210° 330° 210° 330° 210° 330°

240° 300° 240° 300° 240° 300° 240° 300° 240° 300°
270° 270° 270° 270° 270°

Horiz. plane Horiz. plane Horiz. plane Horiz. plane Horiz. plane
(294 Hz) (588 Hz) (882 Hz) (1176 Hz) (2058 Hz)
dB
90 90 90 90 90 0
elevation θ [deg]

-5

60 60 60 60 60 -10

-15

30 30 30 30 30 -20

-25

0 0 0 0 0 -30
-180 -90 0 90 180 -180 -90 0 90 180 -180 -90 0 90 180 -180 -90 0 90 180
0 -180 -90 0 90 180
azimuth φ [deg] azimuth φ [deg] azimuth φ [deg] azimuth φ [deg] azimuth φ [deg]
Full pattern Full pattern Full pattern Full pattern Full pattern
(294 Hz) (588 Hz) (882 Hz) (1176 Hz) (2058 Hz)

Fig. 4: 3D radiation pattern of the violin under analysis. Results in the horizontal plane are compared with the reference data
(dashed line) obtained in a controlled scenario.

musician to play in a more natural fashion, without forcing [8] A. Perez Carrillo, J. Bonada, J. Patynen, and V. Val-
unnatural poses or playing in anechoic rooms. Moreover, the imaki. Method for measuring violin sound radiation
measurement also includes the effect of the head and body of based on bowed glissandi and its application to sound
the player, thus becoming more realistic. synthesis. The Journal of the Acoustical Society of
America, 2011.
REFERENCES [9] D. Markovic, F. Antonacci, A. Sarti, and S. Tubaro.
[1] G. Bissinger. Some mechanical and acoustical conse- Soundfield imaging in the ray space. IEEE Transactions
quences of the violin soundpost. The Journal of the on Audio, Speech, and Language Processing, 2013.
Acoustical Society of America, 1995. [10] E. G. Williams. Fourier Acoustics - Sound radiation
[2] G. Weinrich. Directional tone color. The Journal of the and nearfield acoustical holography. Academic Press,
Acoustical Society of America, 1997. 1999.
[3] G. Bissinger and J. Keiffer. Radiation damping, effi- [11] H. G. Kim, N. Moreau, and T. Sikora. MPEG-7 au-
ciency, and directivity for violin normal modes below 4 dio and beyond: audio content indexing and retrieval.
khz. The Journal of the Acoustical Society of America, Wiley, 2005.
2002. [12] S. O. H. Madgwick, A. J. L. Harrison, and
[4] L. M. Wang and C. B. Burroughs. Acoustic radiation R. Vaidyanathan. Estimation of imu and marg orien-
from bowed violins. The Journal of the Acoustical So- tation using a gradient descent algorithm. In IEEE
ciety of America, 2001. International Conference on Rehabilitation Robotics,
ICORR, 2011.
[5] E. Ravina, P. Silvestri, P. Montanari, and G. De Vec-
chi. Spherical mapping of violins. The Journal of the [13] P. Stoica and R. L. Moses. Introduction to Spectral
Acoustical Society of America, 2008. Analysis. Prentice Hall, 1997.
[6] J. Pätynen and T. Lokki. Directivities of symphony or- [14] R. Moses P. Stoica. Introduction to Spectral Analysis.
chestra instruments. Acta Acustica united with Acus- Upper Saddle River, NJ: Prentice Hall, 1997.
tica, 2010. [15] J. Meyer. Acoustics and the performance of music:
[7] D. F. Comesana, T. Takeuchi, S. M. Cervera, and K. R. manual for acousticians, audio engineers, musicians,
Holland. Measuring musical instruments directivity architects and musical instruments makers. Springer,
patterns with scanning techniques. In International PPV Medien, 2009.
Congress on Sound and Vibration, ICSV, 2012.

Authorized licensed use limited to: Petrobras. Downloaded on 1550

July 16,2020 at 13:23:22 UTC from IEEE Xplore. Restrictions apply.