Quantification of the Voicescape: A Person-centric Approach to

Describing Real-life Behaviour Patterns

A Case Study Comparing Two Age Groups

Ana Londral

1,2

, Burcu Demiray

3,4

and Marcus Cheetham

1,3

Department of Internal Medicine, University Hospital Zurich, Zurich, Switzerland

Institute of Molecular Medicine, University of Lisbon, Portugal

University Research Priority Program “Dynamics of Healthy Aging”,

University of Zurich, Zurich, Switzerland

Division of Gerontopsychology and Gerontology, Department of Psychology,

University of Zurich, Zurich, Switzerland

Keywords: Speech Signal Processing, Voicescape, Person-Centric, Behaviour Analysis, Ageing.

Abstract: The human voice is a fundamental part of the everyday auditory environment. A measure of all voice activity

that a person produces or perceives in the environment, i.e., the person’s voicescape, might provide an

informative, low cost, ecologically valid, and person-centric approach to characterizing patterns of socially-

relevant behaviour in real life. In this paper, we use the measure ratio of voice activity (rva) and present results

of data acquired from N=20 subjects of 2 different age groups as they engaged in their usual daily life activities

over 4 consecutive days. The data show no differences in total voice activity but significant between-group

differences in its daily distribution. We propose that measurement of the voicescape can, even without

knowledge of specific voice sources, serve as a useful indicator of person- or group specific activity patterns

for purposes of describing significant aspects of variation and within- and between-group differences in

patterns of everyday behaviour and, potentially, for identifying change in patterns that have health-related

implications. Future work will target automatic detection and identification of voice sources and the use of

privacy-preserving processing methods.

1 INTRODUCTION

The human voice is a fundamental part of a person’s

everyday auditory environment. The combination of

all voice activities that a person produces or perceives

within the auditory environment may be referred to as

the person’s voice soundscape (cf. Schafer

1994[1977]) or simply voicescape. Even at a low-

level of granularity, the voicescape provides a

potentially informative means to exploring and

characterizing patterns of socially-relevant behaviour

from a person-centric perspective in the natural

setting.

Real-life patterns of behavioural activity are of

longstanding interest (Fahrenberg et al, 2007), more

recently in the field of aging research. Age-related

research of everyday behaviours has focused on

physical activity, with growing interest in mobility,

social context, and time-location patterns (e.g.,

Khusainov et al, 2013). But as the physical and social

environment of the speaker changes due to the impact

of aging on normal functioning in daily life (Wahl &

Lang, 2004), quantitative and qualitative aspects of

the voicescape may be expected to change, too.

In the present paper, we analysed all voice activity

of 2 groups of 20 younger and older healthy subjects

in sound samples acquired unobtrusively using a

wearable device while subjects engaged in their daily

activities. For analysis, we used the feature ratio of

voice-to-nonvoice activity (rva) in each sound sample

as an indicator of activity in the voicescape.

This paper describes the tool that was developed

to extract and quantify voice activity in the samples,

presents the results based on rva, and demonstrates

the visualization tool developed to present the results.

The results show different patterns of activity in the

voicescape of the two age groups.

We suggest that low-level measures of the

voicescape (e.g. rva, sound versus silence, noise

exposure) can play a useful role in describing and

bringing attention to significant aspects of variation

Londral, A., Demiray, B. and Cheetham, M.

Quantiﬁcation of the Voicescape: A Person-centric Approach to Describing Real-life Behaviour Patterns.

DOI: 10.5220/0006653703110314

In Proceedings of the 11th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2018) - Volume 4: BIOSIGNALS, pages 311-314

ISBN: 978-989-758-307-0

311

and within- and between-group differences in cross-

sectional or longitudinal patterns of everyday

behaviour. We discuss the potential application of a

person’s voicescape in studies of health aging and e.g.

for alerting health professionals to unexpected health-

relevant changes in habitual patterns of behavioural

activity that may escape self-report.

2 METHODOLOGY

2.1 Dataset

For this study, we used a data set that was collected

over 4 consecutive days, using the Electronically

Activated Recorder (EAR) (Mehl & Pennemaker,

2003) as a method for naturalistic observation in daily

life. Each participant used a smartphone and app to

randomly activate the microphone from the

smartphone to record 30 seconds of sound (on

average 4 samples/hour). Each 30s sample was stored

as a wav. file. The recordings were active from 6:00

to 00:00. The protocol is otherwise as described e.g.

in Mehl & Pennemaker (2003).

We randomly analysed 20 participants from this

dataset: 10 in the age group of the young (Y) and 10

in the age group of the elderly (E). 60% of the

participants were women. This study was

2.2 Extraction of Voice Activity

From the process of segmentation, the total amount of

voice per each sample was calculated as the sum of

the duration of all the voiced segments. According to

equation (Equation 1), the rva is defined for each

sample as the percentage of voice activity in the total

30-sec sample.

 = 

∑

∆(



)



(Equation 1)

2.3 Grouping and Resampling

For each subject, we re-sampled the data to 2h

periods, and automatically calculated the average

values of rva of all the samples recorded in each of

the following periods: {6:00-8:00; 8:00-10:00;

10:00-12:00; 12:00-14:00; 14:00-16:00; 16:00-

18:00; 18:00-20:00; 20:00-22:00; 22:00-24:00}.

3 RESULTS

A total of 55.7 hours of sound were recorded and

23.07 hours of voice activity detected in the

voicescape of the 20 subjects (described in Table 1).

The rva per sample is represented in Figure 1,

indicating that most samples have either very low

voice activity (rva <2%) or very high voice activity

(rva >98%), independently of the age group.

A visual representation of the rva of two subjects

is presented in this paper as a tool for visual behaviour

analysis. As depicted in Figure 2, the representation

of voice activity for two subjects allows identification

of differences in the voice environment that may be

related to age characteristics. For example, it is clear

that the subject in the group Y has higher voice

activity than the subject in group O, at the late

evening period (from 22:00), but the opposite is

observed in the early morning period (from 8:00).

Table 1: Description of the dataset.

Total of persons 20

Mean age per Age group 69.9±4.7 years old

(O) 23.2±3.0 years

old (Y)

Average amount of sound

recorded per subject

(hours)

3.1 ± 0.9

Average voice detected

per subject (min)

76.9 ± 23.2

Average voice detected

per age group (min)

74.5±27.1 (O)

79.4±18.1 (Y)

Statistical analysis of rva and different periods of the

day indicate significant differences (ANOVA2-

way(time_segment, age group) p<0.001, η

=0.02)

between age groups and time periods. Visual

inspection of Figure 3 indicates a time period in

which there are strong between-group differences in

voice activity at 14:00-16:00 p.m. At this time, voice

activity of subjects in the older group decreases

drastically. On the other hand, rva at late evening is

considerably higher for the younger group. Generally,

it is possible to observe that the rva for the subject in

group O is higher by the end of the morning period

and by beginning of evening period. This fact is

probably related to lunch and dinnertime, when older

subjects may have the most social moments.

Real 2018 - Special Session on Assessing Human Cognitive State in Real-World Environments

312

Figure 1: Histogram containing the rva per sample.

Interestingly, when considering the total rva

independently of time of day, the voice activity is

similar in both groups, as depicted in the histogram of

Figure 1. While this suggests the same overall amount

of voice activity for the subjects independently of the

age group, the sources of voice activity may differ

between groups.

Figure 2: Graphical representation of voice activity of two

subjects: Left panel shows a subject from the older group

(O) and the right panel from the younger group (Y).

Figure 3: Mean rva in each time segment, separated by age

groups. Vertical bars represent the confidence interval

(95%).

Automatic classification of voice sources was not

applied in this study. But manual auditory inspection

was performed in random samples containing high

rva (>90%). This showed that voice activity in older

persons is mostly related to the TV, whereas in

younger subjects voiced backgrounds (e.g. classroom

or restaurant) dominate the voicescape.

4 CONCLUSIONS

In this paper, we propose the use of voice activity in

the voicescape as a potentially informative, low cost,

ecologically valid, person-centric approach to chara-

cterizing patterns of socially-relevant behaviour. In

contrast to most studies in speech processing, where

voice is recorded in controlled environments, this

study used sounds recorded in the self-selected

natural setting of the subjects.

The ratio of voice activity was analysed as the

percentage of voiced segments in 30s-samples over 4

consecutive days to map daily patterns of voice

activity. This paper presents a case study with two age

groups. We observed no overall difference in rva of

younger and older adults in the natural setting and a

common pattern of either extremely low or high rva

in the samples, such that voice activity was either

very low (rva<2%) or very high (rva>90%). Across

the day, however, there were significant differences

in voice activity in two specific time periods. Voice

activity was lower in the older group over the midday

period and late evening.

We conclude that voice activity present in a

person`s soundscape can, even without knowledge of

specific voice sources, serve as an indicator of

person- or group specific behavioural patterns for

purposes of exploring significant areas of further

research interest. This approach might be used to

examine associations between health-related factors

and patterns of habitual social-behavioural activity

and to indicate deviations in habitual patterns of

behaviour that may escape self-report but have

health-related implications. Future work will target

automatic detection and identification of voice

sources. With a focus on the voicescape and voice

sources, this work can be conducted using privacy-

preserving processing methods (e.g., Glackin et al.,

2017).

ACKNOWLEDGEMENTS

This work was funded by the Department of Internal

Medicine, University Hospital Zurich, and the

University Research Priority Program “Dynamics of

Healthy Aging”, University of Zurich, Switzerland.

REFERENCES

Blake, E. C., & Cross, I. (2014). The Acoustic and Auditory

Contexts of Human Behavior. Current Anthropology,

Quantiﬁcation of the Voicescape: A Person-centric Approach to Describing Real-life Behaviour Patterns

313

56 (1), 81-103. The University of Chicago

PressWenner-Gren Foundation for Anthropological

Research

Davies, William J. 2010. The Acoustic Environment.

Oxford University Press.

Fahrenberg, Jochen, Michael Myrtek, Kurt Pawlik,

Meinrad Perrez., 2007. Ambulatory Assessment -

Monitoring Behavior in Daily Life Settings: A

Behavioral-Scientific Challenge for Psychology.

European Journal of Psychological Assessment 23 (4):

206–13.

Glackin. C., Chollet, G., Dugan.N., & Rajaraja, M. (2017).

Privacy preserving encrypted phonetic search of speech

data," 2017 IEEE International Conference on

Acoustics, Speech and Signal Processing (ICASSP),

New Orleans, LA, pp. 6414-6418.

Ludlow, Christy L., 2015. Central Nervous System Control

of Voice and Swallowing. Journal of Clinical

Neurophysiology : Official Publication of the American

Electroencephalographic Society 32 (4). NIH Public

Access: 294–303.

Khusainov, R., Azzi, D., Achumba, I. E., & Bersch, S. D.

(2013). Real-Time Human Ambulation, Activity, and

Physiological Monitoring: Taxonomy of Issues,

Techniques, Applications, Challenges and Limitations.

Sensors (Basel, Switzerland), 13(10), 12852–12902.

Mehl MR, Pennebaker JW (2003) The sounds of social life:

A psychometric analysis of students' daily social

environments and natural conversations. J Pers Soc

Psychol., 84(4):857-70.

Muaremi, A., Bert Arnrich, and Gerhard Tröster. 2013.

‘Towards Measuring Stress with Smartphones and

Wearable Devices During Workday and Sleep’.

BioNanoScience 3 (2): 172–83.

Narayanan, S., Panayiotis G., 2013. ‘Behavioral Signal

Processing: Deriving Human Behavioral Informatics

from Speech and Language’. Proceedings of the IEEE

101 (5): 1203–33.

Schafer, R. Murray. 1994 (1977). The Soundscape: Our

Sonic Environment and the Tuning of the World.

Rochester,. Vermont: Destiny, 1994 (1977). pp. 293.

Wahl, H.-W. & Lang, F. (2004). Aging in context across

the adult life course: Integrating physical and social

environmental research perspectives. In H.-W. Wahl,

R. Scheidt & P. Windley (eds.), Aging in Context:

Socio-Physical Environments (Annual Review of

Gerontology and Geriatrics 23) (pp. 1–34). New York:

Springer.

Real 2018 - Special Session on Assessing Human Cognitive State in Real-World Environments

314