Assessing the QoME of NMP via Audio Analysis Tools

Konstantinos Tsioutas

and George Xylomenos

Department of Informatics, Athens University of Economics and Business, Athens, Greece

Keywords:

NMP, QoME, Tempo Analysis.

Abstract:

Analyzing the Quality of Musicians’ Experience (QoME) in Network Music Performance (NMP) typically

involves having musicians perform NMP sessions and then assessing their experience via questionnaires. Such

subjective studies produce results with wide variances, making the extraction of solid conclusions difﬁcult.

For this reason, we complemented a subjective study on the effects of delay in the QoME of NMP with an

analysis of the audio captured during the study using automated tools. Speciﬁcally, we used signal processing

techniques to analyze the captured audio, in order to detect tempo evolution during each performance and

examine its correlation with delay. Our results indicate that musicians in real NMP settings are more tolerant

to delay than previously thought, holding a steady tempo even with one way delays of 40 ms.

1 INTRODUCTION

The Quality of Musicians’ Experience (QoME) in

Network Music Performance (NMP), that is, the per-

formance of music when musicians are connected

over a network, is a complex function which depends

on many variables, including audio delay and audio

quality (Tsioutas et al., 2020). As in other human-to-

human communication applications, NMP has strict

delay requirements: while regular video conferenc-

ing can tolerate up to 100 ms of one way delay, in

NMP delays of more than 20–30 ms are considered

problematic; this delay limit is called the Ensemble

Performance Threshold (EPT) (Schuett, 2002).

Such delays are hard to achieve over the wide area,

even with very high speed networks, as increasing

transmission speed only reduces transmission delay;

propagation delay depends on the distance travelled,

while queueing delay depends on router load and the

number of hops in the network path. It is even harder

to achieve such delays with residential DSL connec-

tions, as many musicians found out during the recent

CoVid-19 pandemic.

Although numerous studies have investigated the

effect of delay in the ability of musicians to synchro-

nize, in an attempt to determine the tolerance of NMP

to delay, there are many pieces missing to understand

the big picture of QoME in NMP. Apparently, QoME

is strongly connected with the perception of various

https://orcid.org/0000-0003-1136-3803

https://orcid.org/0000-0003-2552-687X

audio phenomena and affected by audio features such

as music rhythm, music tempo and audio spectral fea-

tures (Rottondi et al., 2015). However, most studies

with actual musicians had a small number of partici-

pants (8 – 12), hence it is hard to derive reliable con-

clusions from them.

As part of our work on the subjective evaluation

of the effects of audio delay on the QoME of NMP,

we have performed a large number of controlled ex-

periments, where pairs of musicians play a musical

piece under different conditions, completing a ques-

tionnaire at the end of each performance (Tsioutas

et al., 2021). The analysis of these questionnaires re-

veals that not only different musicians perceive the

same conditions in quite different ways, even the re-

sponses from the same musicians are not consistent

with the underlying parameters; for example, their as-

sessment of delay does not follow the actual delay

in the experiments. The results from these subjective

evaluations thus exhibit a very high variance, which

makes drawing concrete conclusions harder.

For this reason, in this study we have chosen to

follow a different path. Having recorded audio (and

video) from our NMP experiments, we decided to

use automated tools to extract information related to

QoME, thus complementing the subjective question-

naires. Speciﬁcally, we used a signal analysis toolkit

to determine how the performance tempo varies as de-

lay is increased, a phenomenon often observed in pre-

vious studies where the tempo tended to slow down

as delay grew; we applied this to actual music perfor-

Tsioutas, K. and Xylomenos, G.

Assessing the QoME of NMP via Audio Analysis Tools.

DOI: 10.5220/0010604100230030

In Proceedings of the 18th International Conference on Signal Processing and Multimedia Applications (SIGMAP 2021), pages 23-30

ISBN: 978-989-758-525-8

mances however, not simple hand clapping synchro-

nization exercises.

The outline of the rest of the paper is as follows.

In Section 2, we brieﬂy present related work on as-

sessing the effects of delay on QoME and the perfor-

mance tempo, in particular. Section 3 describes the

technical setup of our experimental scenarios and the

measurement process. In Section 4 we present a quan-

titative analysis of the NMP sessions that shows how

tempo is affected by delay during the performances.

We summarize our ﬁndings and discuss future work

in Section 5.

2 RELATED WORK

A large amount of research touches upon QoME eval-

uation for NMP, looking at it from different perspec-

tives. Most studies are subjective, that is, the partici-

pants perform one or more experimental sessions and

then answer surveys evaluating their experience while

audio delay or other variables are manipulated.

The pioneer study of (Schuett, 2002) investigated

the effects of delay to synchronization, proposing the

term Ensemble Performance Threshold (EPT) for the

one way delay below which synchronization is possi-

ble, reporting that it lies between 20 and 30 ms. An-

other study reached the same conclusion, experiment-

ing with musicians who clapped their hands, indicat-

ing that when the latency was below 11 ms, the tempo

accelerated (Chafe and Gurevich, 2004).

Conversely, an experiment with two musicians

who performed a clapping session without any ex-

ternal tempo reference (i.e., there was no starting

metronome or countdown) found that the tempo of

the hand claps slowed down as the delay was in-

creased (Driessen et al., 2011).

In a study of eleven pairs of musically experi-

enced subjects that attempted to synchronize their

hand claps, the tempo was found to decrease more

rapidly for higher delays, and the relation between

delay and tempo slowdown was found to be approxi-

mately linear (Farner et al., 2009).

Finally, in an experiment where seventeen pairs of

subjects performed hand clapping sessions under vari-

able time delays, the authors reported that for delays

shorter than 11.5 ms, 74% of the performances sped

up, while at delays of 14 ms and above, 85% of the

performances slowed down. No correlation with the

starting tempo was found in the range sampled (Gure-

vich et al., 2004).

Turning to actual musical performances, in (Bar-

bosa et al., 2005) four musicians played bass, percus-

sion, piano and guitar. The authors found that regard-

less of the instrumental skills or the musical instru-

ment, all musicians were able to tolerate more delay

when slower tempos were used.

Another experiment investigated how the attack

period of notes (that is, the time in which a note

reaches it maximum amplitude) affects the tempo, us-

ing two musicians performing cello and violin. The

analysis of the recordings showed that the tempo was

generally higher with fast attack times than with slow

attack times. In both cases, it decreased with increas-

ing latency (Barbosa and Cordeiro, 2011).

Using a conductor is common in orchestral set-

tings, therefore in (Olmos et al., 2009) six singers and

a conductor performed over a network. For the most

part, the singers that participated in the experiment

managed to cope with the various delays. The singers

mentioned that to a certain extent, they were able to

establish emotional connections with each other.

In (Bartlette et al., 2006) two pairs of musicians

were asked to performed two Mozart duets. Although

the musicians chose different strategies to handle the

latency, both duets were strongly affected by latency

at and above 100 ms. At these levels, the musicians

rated the performances as neither musical nor interac-

tive, and they reported that they played as individuals

and listened less and less to one another.

In a very comprehensive study of NMP (Rottondi

et al., 2015), the authors asked seven pairs of musi-

cians to participate in the experiments. Each repeti-

tion was characterized by different settings in terms of

reference tempo, network latency, and jitter. The au-

thors reported that the perceived delay was strongly

affected by the timbral and rhythmic characteristics

of the combination of instruments and parts. They

also noted that the noisiness of the instrument has an

impact on the perceived delay, for example, perform-

ers of percussive instruments reacted more strongly

to delay changes. They concluded that the possibility

of enjoying an NMP session is not only a function of

delay, but also of the role and the timbral characteris-

tics of the involved musical instruments, as well as the

rhythmic complexity of the musical pieces performed.

In (Car

ot et al., 2009) ﬁve professional drum-

mers performed with ﬁve professional bass players.

The authors reported that the overall delay thresholds

ranges between a minimal delay of 5 ms and a maxi-

mal delay of 65 ms. They also noted that the players

did not exhibit a common latency acceptance value,

indicating that tolerance to delay is a subjective issue.

Finally, in (Delle Monache et al., 2018) ten volun-

teers participated in pairs, performing mandolin, ac-

cordion, guitar, percussion, harp, ﬂute, alto sax. De-

lay had a negative effect to the involvement of the mu-

sicians with the process. The authors also reported

SIGMAP 2021 - 18th International Conference on Signal Processing and Multimedia Applications

Figure 1: Experimental Setup.

that a general distress was caused by latency, but a

willingness to ﬁnd ways to cope with it also emerged

from the answers.

To summarize, the hand clapping studies indi-

cate that synchronization is hard when delay exceeds

30 ms, with the tempo slowing down as delay rises be-

yond this threshold. However, studies with real mu-

sical performances show quite diverse results, indi-

cating that in real NMP sessions musicians do adapt

to higher delays, often by slowing down their tempo,

depending on the type of music performed and the in-

struments used.

3 EXPERIMENTAL SETUP

For our experiments, we used two visually and au-

rally isolated rooms on the same ﬂoor of the main

AUEB building. Musicians performed with their

counterparts in separate rooms, while listening to

them through headphones and seeing them through a

32” video monitor.

As shown in Figure 1, an eight channel mixing

console was used in each room for the necessary au-

dio routing, monitoring and recording. Audio was

captured by condenser microphones and closed type

headphones were used by the musicians to listen to

each other. A video camera was capturing and send-

ing a composite video signal through the existing

network cabling to the video monitor of the other

room (red lines in the ﬁgure). The network cables

were patched directly to each other, without passing

through any network equipment; we basically used

one pair of the UTP cables to transmit the compos-

ite video signal in analog mode. We used composite

video in order to achieve the lowest possible visual

delay between musicians; with the analog signal we

did not have to wait for entire frames to be captured

before transmission and received before display.

We experimentally measured the round trip delay

by placing a smartphone with a running chronometer

in front of the camera in one room, and turning the

video camera to the video monitor in the other room,

essentially reﬂecting the transmitted image back to

Figure 2: My Mouth to My Ear delay.

the ﬁrst room. We then recorded with another smart-

phone’s camera both the chronometer and its reﬂected

image, and analyzed the video in a video editor, ﬁnd-

ing out that the round trip delay was 30 ms, therefore

the one way delay was 15 ms.

The two mixing consoles were also connected

through the existing network cabling, using direct ca-

ble patching, hence the audio signal was also trans-

mitted in analog form from one room to the other.

The reason for connecting them directly was to be

able to achieve perfectly ﬁxed audio delays even be-

low 10 ms, which is impossible when computers and

network devices intervene in the signal path; the prop-

agation delays were less than 1 ms due to the small

cabling distance between the rooms. To manipulate

audio delay we used AD-340 audio delay boxes by

Audio Research, via which we were able to set delay

in each direction to the desired value. We attached

a PC with a Motu 828X external audio interface (not

shown) to the auxiliary output of the mixing console

to record each session for later analysis, without in-

troducing delays in the signal path.

Unlike most NMP studies which use Mouth to

Ear (M2E) delay, which is the end-to-end delay be-

tween the microphone at one end and the speaker at

the other end, in our work we use the My Mouth to

My Ear (MM2ME) delay proposed in (Tsioutas et al.,

2020). As shown in Figure 2, MM2ME is the two-

way counterpart to M2E, over which it has three ad-

vantages. First, when musicians play together, each

musician plays one note and unconsciously expects to

listen to the other musicians’ note to play his next one,

and so on. Second, measuring MM2ME delay accu-

rately is much easier than measuring the M2E delay,

as it can be done at one endpoint, by simply reﬂecting

the transmitted sound at the other endpoint; in con-

trast, M2E needs to be measured at both endpoints,

thus requiring perfectly synchronized clocks, some-

thing very hard to achieve (Car

ot et al., 2020). Third,

MM2ME takes into account the possible asymmetry

between the two directions of a connection, which is

Assessing the QoME of NMP via Audio Analysis Tools

Table 1: Instruments played by the musicians.

Duet 1 Duet 2 Duet 3 Duet 4 Duet 5 Duet 6 Duet 7 Duet 8 Duet 9 Duet 10 Duet 11

Folk Folk Rock Rock Funk Funk Rock Rock Classic Folk Folk

Piano Piano El Gtr El bass Organ El bass El bass El Gtr Flute Ac Gtr Laoud

Santuri Oud El Gtr El Gtr El Gtr Perc Ac Gtr Violin Violin Buzuki Violin

Table 2: MM2ME delays in order of use.

Repetition 1 2 3 4 5 6 7 8 9 10

MM2ME delay (ms) 10 25 35 30 20 0 40 60 80 120

the rule with residential DSL endpoints.

We conducted experimental sessions with 22 indi-

vidual musicians (11 duets); to the best of our knowl-

edge, this is the largest NMP study with actual musi-

cal performances (as opposed to hand claps) to date.

The musicians performed with a variety of instru-

ments, including piano, acoustic guitar, electric gui-

tar, electric bass, violin and ﬂute, as well as traditional

instruments including the lute, toumberleki, santuri

and oud, in a musical style of the choice. Table 1

shows the musical style and the instruments for each

duet. Each pair of musicians played a musical part of

their choice, with a duration of up to 60 sec, follow-

ing their own tempo and repeating it ten (10) times,

using a different MM2ME delay setting for each rep-

etition. We kept the duration low, to avoid tiring the

musicians, since they had to repeat the piece multiple

times and subjectively assess their experience in the

end. Table 2 shows the delays used and the sequence

with which they were applied to each repetition; half

of that delay was set in each direction via the audio

delay device. No metronome or other synchroniza-

tion aids were used.

Note that a delay of 0 ms is unnaturally low: two

musicians 2 meters apart from each other experience a

one way delay of 5.83 ms (11.66 ms MM2ME) based

on the speed with which sound propagates through the

atmosphere; a duet in the same room would therefore

experience a natural MM2ME delay of 10 to 20 ms,

depending on their positions in the room.

After the end of each repetition, each musician

was asked to answer an electronic questionnaire on

a tablet (see (Tsioutas et al., 2021) for details). Mu-

sicians were not informed about which variable was

manipulated each time, or about the purpose of the ex-

periment, and we randomly set the order in which the

audio delay values were set for each repetition. The

main goal was to conduct an experiment that would

allow us to evaluate multiple variables without bias or

noise in the answers. The audio (and video) of each

performance was recorded, and was used for the anal-

ysis discussed in the following sections.

4 TEMPO ANALYSIS

As mentioned in Section 2, experiments where par-

ticipants tried to synchronize their hand claps over

the network have indicated that as the delay between

the endpoints grows, the participants compensate by

slowing down their tempo. Since hand claps have a

simple audio signature, it is easy to note such slow-

downs by simply looking at the waveform of the

recordings. The same observation was made in some

experiments with real musicians, even though the ex-

act correlation between the delay and the tempo was

harder to quantify with the more complex sonic im-

print of actual musical pieces.

Although our original goal when setting up an

NMP study was to perform a subjective evaluation of

the effect of delay on QoME via questionnaires, we

had recorded the audio tracks of each session for later

analysis. This allowed us to assess whether the tempo

does indeed grow as delay is increased in our more

realistic setup, with actual musicians playing real in-

struments and real musical pieces. Of course, since

each duet selected their own musical piece and tempo,

we had to recover all relevant information from the

actual recordings. That is, unlike in (Rottondi et al.,

2015), we did not know what the intended tempo of

each performance was.

To this end, we analysed the audio recordings us-

ing the MIRToolbox (Lartillot et al., 2013). To deter-

mine the tempo at a period of time, we start with the

event density, which estimates the average number of

note onsets per second as follows:

E =

(1)

where E is event density, O is the number of note on-

sets and T is the duration of the musical piece. The

MIRToolbox estimates how the music tempo, mea-

sured in Beats per Minute (BPM), varies over time,

by detecting the note onsets via signal processing of

the audio. The analysis is not perfect, as it depends on

each instrument’s sonic signature and manner of play-

ing, but it is revealing, especially for instruments with

SIGMAP 2021 - 18th International Conference on Signal Processing and Multimedia Applications

Figure 3: Tempo variation over time for various delay val-

ues: Duet 1, Piano-Rhythm-Folk.

Figure 4: Tempo variation over time for various delay val-

ues: Duet 1, Santouri-Solo-Folk.

very clear sonic signatures, for example percussion,

or with performances where the instrument plays a

rhythmic pattern. We performed this analysis for the

audio recording of each side of an NMP session.

These results are not easily amenable to numeri-

cal summarization, since musicians adapt their play-

ing over time as they listen to each other; as a result,

each performance leaves a unique time-varying im-

print. However, when presented visually, they show

interesting trends. The following ﬁgures show how

the tempo (in BPM) varies over time (in seconds) for

different musical instruments; each ﬁgure shows one

such curve for each delay value, corresponding to one

performance by a single musician, with progressively

lighter curves corresponding to increasing MM2ME

delays. To reduce visual clutter, we only show results

at 40 ms intervals, that is, with 0, 40, 80 and 120 ms

MM2ME delays.

Figures 3 and 4 show the delay variation for each

Figure 5: Tempo variation over time for various delay val-

ues: Duet 2, Piano-Rhythm-Folk.

Figure 6: Tempo variation over time for various delay val-

ues: Duet 2, Oud-Solo-Folk.

Figure 7: Tempo variation over time for various delay val-

ues: Duet 3, Electric Guitar-Solo-Rock.

instrument of duet 1. We can see that with a delay of

0 ms, which is unnaturally low, as explained above,

both musicians actually speed up their tempo in the

ﬁrst part of the performance, as reported in previous

studies. As the delay grows, the tempo slows down,

but the musicians have a hard time keeping a steady

tempo at all delay values, as evidenced from the ups

and downs in the curves.

While in duet 1 the musicians have trouble keep-

ing a steady rhythm, in duet 2, Figures 5 and 6 show a

different situation: the instrument playing the rhythm

part, in this case the piano, is visibly affected by delay,

since as the delay grows, the tempo drops; however,

the tempo is steady in all but the highest delay value.

The instrument playing the solo part however, in this

case the oud, shows larger tempo variations, even

though the tempo does generally drop with growing

delay. Of course, due to the method we are using to

detect the tempo (note onsets), solo parts where mu-

Figure 8: Tempo variation over time for various delay val-

ues: Duet 3, Electric Guitar-Solo-Rock.

Assessing the QoME of NMP via Audio Analysis Tools

Figure 9: Tempo variation over time for various delay val-

ues: Duet 5, Organ-Solo-Funk.

Figure 10: Tempo variation over time for various delay val-

ues: Duet 6, Percussion-Rhythm-Rock.

sicians play more freely and improvise are harder to

characterize precisely in terms of tempo.

In duet 3 where both musicians have a solo role,

we can see in Figures 7 and 8 that both exhibit tempo

variations, however, the musicians do manage to keep

a relatively steady tempo, except for the highest delay

value of 120 ms. Again, the tempo tends to drop with

higher delays. Note that since both musicians have

improvisational roles, they end their performance at

different time points for each delay value - they do

ﬁnish at the same time, though.

The difﬁculty of keeping a steady tempo at higher

delays is also apparent in Figure 9 which shows the

solo instrument of duet 5 (organ); again, tempo drops

with higher delays, and has wild variations at a de-

lay of 120 ms. With the percussion instrument of

duet 6, arguably the most rhythmic of instruments and

the easiest in terms of automated tempo detection, as

shown in Figure 10, the beat is noticeably slower for

Figure 11: Tempo variation over time for various delay val-

ues: Duet 7, Bass-Rhythm-Rock.

Figure 12: Tempo variation over time for various delay val-

ues: Duet 7, Acoustic Guitar-Solo-Rock.

Figure 13: Tempo variation over time for various delay val-

ues: Duet 8, Guitar-Rhythm-Rock.

higher delays, and hard to keep steady when delay

reaches 120 ms.

There are also cases where both sides of a duet

can keep the same rhythm, as with duet 7, shown in

Figures 11 and 12: the rhythm is steady with delays of

up to 80 ms; there is a very slight reduction in tempo

from 40 to 80 ms, but at 120 ms the tempo either slows

down continuously or varies wildly.

Duet 8 is unusual, in that the rhythm instrument

(guitar), shown in Figure 13 has an unsteady tempo,

while the solo instrument (violin), shown in Figure 14

has a very steady tempo, despite the visible slow-

down at delays of 80 and 120 ms. The reason for

this is the very different expertise levels of the mu-

sicians: the violinist was a 45 year old professional

musician, while the guitarist was a 23 year old ama-

teur one. Hence the violinist’s solo tempo was found

to be more stable than the guitarist’s, even though it

was the guitarist who was supposed to keep a stable

Figure 14: Tempo variation over time for various delay val-

ues: Duet 8, Violin-Solo-Rock.

SIGMAP 2021 - 18th International Conference on Signal Processing and Multimedia Applications

Figure 15: Tempo variation over time for various delay val-

ues: Duet 10, Guitar-Rhythm-Folk.

Figure 16: Tempo variation over time for various delay val-

ues: Duet 11, Lute-Rhythm-Folk.

rhythm with the guitar. This is an indication that more

experienced musicians may manage to partially com-

pensate for delay by adapting their performance.

Finally, the rhythm instruments of duet 10 and

duet 11, shown in Figures 15 and 16 further verify the

previous observations of tempo speedup at the unusu-

ally low delay of 0 ms, good tempo stability at 40 and

80 ms, albeit at a slight reduction of tempo at 80 ms,

and higher variations at 120 ms.

From these ﬁgures, we can make the following

general observations:

1. At the (unnaturally) low delay of 0 ms, musicians

tend to speed up their tempo in the beginning of

the session.

2. As delays rise beyond 40 ms, musicians adapt by

slowing down the tempo of their performance.

3. Instruments performing rhythmic parts are more

clearly affected by delay, as shown by their more

visibly delineated curves.

4. Percussion instruments, which generally have a

rhythmic role, are the most sensitive to delay.

5. In most cases, musicians manage to keep a steady

tempo at delays of up to 80 ms.

6. At a delay of 120 ms performances break down,

exhibiting either continuously slowing or wildly

varying tempos.

These observations verify ﬁndings from past work

that musicians who perform percussive instruments

suffer more from delay than others. Indeed, the hand

clap experiments, where the rhythmic patterns are

very clear, fall in the same category. Of course, these

instruments, with their easy to detect sonic signatures

and their clear temporal patterns, are ideal for this

type of analysis. We can further observe that this is

true for instruments having a rhythmic role in a duet.

Although solo instruments seem to follow more irreg-

ular tempos, we must keep in mind that this may be an

artefact of our audio analysis which relies on a steady

production of note onsets; with improvisational parts,

performers are expected to more often deviate from

the base rhythmic pattern.

The most interesting observation however is that

the limits to tolerance can vary considerably; most

musicians could achieve a stable tempo at MM2ME

delays of 80 ms, corresponding to an one way delay

of 40 rather than 20–30 ms, higher than what was pre-

viously considered the limit to synchronization, even

though this may come at the the cost of a minor slow

down in the performing tempo.

Finally, it should also be noted that we performed

an ANOVA analysis for repeated measures of the av-

erage tempo scores for each session and for delays of

0, 40, 80 and 120 ms (MM2ME) and the p value was

computed equal to 0.007 (p < 0.05). This indicates a

strong statistical signiﬁcance in the delay/tempo rela-

tionship, that is, the calculated tempos were found to

be statistically correlated with the delay values, that

is, higher delays did lead to slower tempos.

5 CONCLUSIONS AND FUTURE

WORK

We conducted a set of NMP experiments, where the

delay between a pair of musicians was varied in a con-

trolled manner for each session, with the audio and

video from the sessions being recorded for later anal-

ysis. In the experiments reported in this paper, 22 mu-

sicians participated as pairs, playing a diverse set of

musical instruments in different musical styles.

The analysis performed on the recorded audio in-

dicates that musicians tend to slow down their tempo

as delays grow, an effect made very clear with per-

cussive instruments and quite clear with instruments

playing rhythmic parts. However, in most cases they

can synchronize and maintain a stable tempo with

MM2ME delays of up to 80 ms (equivalent to 40 ms

one way delays), indicating that the delay tolerance

of actual musicians performing in NMP scenarios is

higher than previously thought, that is, the EPT is

closer to 40 rather than 20–30 ms. Indeed, musi-

cians, especially more experience ones, try to adapt to

Assessing the QoME of NMP via Audio Analysis Tools

higher delays by slowing down their tempo. This con-

clusion is also supported by the analysis of the QoME

questionnaires, reported in (Tsioutas et al., 2021).

Our work continues with a deeper analysis of

the audio data collected, focusing on issues such as

the dependence of tempo variations on other factors,

such as the style of music performed and the target

tempo of each piece. By grouping similar perfor-

mances together, we hope to be able to derive quan-

titative expressions of the relationship between delay

and tempo, depending on those factors.

Similarly, we are currently analyzing the video

data gathered during the sessions via machine learn-

ing techniques, and speciﬁcally facial emotion recog-

nition, in an attempt to quantify the emotional re-

sponse of the participants in an NMP session to delay.

ACKNOWLEDGEMENTS

We would like to thank all the participating musicians

for their patience during the experiments, as well as

the fellows who helped with setting up and carrying

out the experiments. We would also like to thank our

colleague Dr. Ioannis Doumanis (Lecturer at the Uni-

versity of Central Lancashire) for help in designing

the experiments.

REFERENCES

Barbosa, A., Cardoso, J., and Geiger, G. (2005). Network

latency adaptive tempo in the public sound objects

system. In International Conference on New Inter-

faces for Musical Expression, pages 184–187.

Barbosa, A. and Cordeiro, J. (2011). The inﬂuence of per-

ceptual attack times in networked music performance.

Audio Engineering Society International Conference.

Bartlette, C., Headlam, D., Bocko, M., and Velikic, G.

(2006). Effect of network latency on interactive mu-

sical performance. Music Perception: An Interdisci-

plinary Journal, 24(1):49–62.

Car

ot, A., Hoene, C., Busse, H., and Kuhr, C. (2020).

Results of the fast-music project – ﬁve contributions

to the domain of distributed music. IEEE Access,

8:47925–47951.

Car

ot, A., Werner, C., and Fischinger, T. (2009). Towards a

comprehensive cognitive analysis of delay-inﬂuenced

rhythmical interaction. In International Computer

Music Conference.

Chafe, C. and Gurevich, M. (2004). Network time delay and

ensemble accuracy: Effects of latency, asymmetry. In

Audio Engineering Society Convention 117.

Delle Monache, S., Buccoli, M., Comanducci, L., Sarti, A.,

Cospito, G., Pietrocola, E., and Berbenni, F. (2018).

Time is not on my side: Network latency, presence

and performance in remote music interaction. In Col-

loquium on Musical Informatics (CIM), pages 20–23.

Driessen, P. F., Darcie, T. E., and Pillay, B. (2011). The

effects of network delay on tempo in musical perfor-

mance. Computer Music Journal, 35(1):76–89.

Farner, S., Solvang, A., Sæbø, A., and Svensson, U. P.

(2009). Ensemble hand-clapping experiments under

the inﬂuence of delay and various acoustic environ-

ments. Journal of the Audio Engineering Society,

57(12):1028–1041.

Gurevich, M., Chafe, C., Leslie, G., and Tyan, S. (2004).

Simulation of networked ensemble performance with

varying time delays: Characterization of ensemble ac-

curacy. In International Computer Music Conference.

Lartillot, O., Cereghetti, D., Eliard, K., Trost, W. J., Rappaz,

M.-A., and Grandjean, D. (2013). Estimating tempo

and metrical features by tracking the whole metrical

hierarchy. In The 3rd International Conference on

Music & Emotion, Jyv

askyl

a, Finland. University of

Jyv

askyl

a, Department of Music.

Olmos, A., Brul

e, M., Bouillot, N., Benovoy, M., Blum, J.,

Sun, H., Lund, N. W., and Cooperstock, J. R. (2009).

Exploring the role of latency and orchestra placement

on the networked performance of a distributed opera.

In 12th Annual International Workshop on Presence,

pages 1–9, USA. ISRP.

Rottondi, C., Buccoli, M., Zanoni, M., Garao, D., Verticale,

G., and Sarti, A. (2015). Feature-based analysis of

the effects of packet delay on networked musical in-

teractions. Journal of the Audio Engineering Society,

63:864–875.

Schuett, N. (2002). The effects of latency on ensemble per-

formance. Bachelor Thesis, CCRMA Department of

Music, Stanford University.

Tsioutas, K., Xylomenos, G., and Doumanis, I. (2021). An

empirical evaluation of QoME for NMP. In IFIP In-

ternational Conference on New Technologies, Mobil-

ity and Security.

Tsioutas, K., Xylomenos, G., Doumanis, I., and Angelou,

C. (2020). Quality of musicians’ experience in net-

work music performance: A subjective evaluation. In

Audio Engineering Society Convention 148.

SIGMAP 2021 - 18th International Conference on Signal Processing and Multimedia Applications