A Computer-based Framework to Process Audiometric Signals using

the Tomatis Listening Test System

Félix Buendía-García

, Manuel Agustí-Melchor

, Cristina Pérez-Guillot

, Hernán Cerna

and Alvaro Capitán

Computing Engineering Department, Universitat Politècnica de València, Camino de Vera s/n, Valencia, Spain

Applied Linguistics Department, Universitat Politècnica de València, Camino de Vera s/n, Valencia, Spain

Isora Solutions, Calle de la Luna, 24, 28004 Madrid, Spain

Keywords: Audiometric Signals, Tomatis Listening Test System, Image and Spreadsheet Data Processing.

Abstract: Some kinds of audio information are usually represented by images that need to be processed. This is the

case of audiometric signals which are obtained from some devices that hardly produce quantifiable data.

The current paper describes a computer-based framework able to process audiometer images in order to

extract information which can be useful to analyse subject's hearing levels. Such information is

complemented with additional data sources that allow a more comprehensive view of hearing issues either

disorder symptoms or treatment results. These data sources are provided by the TLTS (Tomatis Listening

Test System) device. The proposed framework is based on the use of OpenCV libraries that provide image

processing functionalities together with scripts to manage audiometry spreadsheets. An experiment has been

developed to test auditory stimulations in the context of a collaboration project with the Isora Solutions

company where the proposed system was applied. Obtained results show the framework accuracy and

adequacy to retrieve and process information from several audiometric data sources.

1 INTRODUCTION

Audiometry deals with the need of measuring

hearing acuity for variations in several audio

parameters. Such audiometric process can be based

on objective measurements coming from physical or

acoustic signals, or relying on subjective user

responses. This second scenario represents the work

context of this paper that involves several signal

sources when determining the subject's hearing

levels. Audiometer devices used for evaluating this

hearing acuity are usually based on embedded

hardware units which barely produce charts called

audiograms as a result of the audiometric process.

The current work focuses on the processing of

the images and spreadsheets that record the

audiogram information as data sources provided by

the Tomatis Listening Test System (TLTS).

Moreover, there are several types of tests that can be

addressed in this context such as Pure Tone

Audiometry (PTA), Masking Level Difference

(MLD) or Speech audiometry that requires different

data types to be processed. The computer-based

framework proposed in this work intends to combine

and integrate those data sources that enable a more

comprehensive and holistic view of audiometry

outcomes in a context of auditory stimulations to

improve listening skills.

Computers are quite useful to help the hearing

assessment when several multimedia signal sources

are combined. For example, Mackersie et al., (2001)

presented the evaluation of speech perception

through a computers-assisted test called CASPA

(Computer Assisted Speech Perception Assessment).

This work was extended in the system CasperSent

(Boothroyd, 2006) as a multimedia program whose

main purpose was sentence-level speech-perception

training and testing.

A set of auditory assessment tests based on

integrating phonetic discrimination and word

recognition were described (Eisenberg et al., 2007).

Fernandez et al., (2014) used detection of eye

gesture reactions as a response to sounds in order to

provide computer aided hearing assessment.

Therefore, a combination of multiple data sources is

essential to get a global view of the specific auditory

scenario. In this sense, processing audiogram charts

Buendía-García, F., Agustí-Melchor, M., Pérez-Guillot, C., Cerna, H. and Capitán, A.

A Computer-based Framework to Process Audiometric Signals using the Tomatis Listening Test System.

DOI: 10.5220/0006431400250034

In Proceedings of the 14th International Joint Conference on e-Business and Telecommunications (ICETE 2017) - Volume 5: SIGMAP, pages 25-34

ISBN: 978-989-758-260-8

is a key issue in the current work together with the

collection of spreadsheets data provided by

audiometers.

Another aspect is about the intervention of

experts in the audiometric process. Automated tests

have been developed to check hearing issues in

specific audiometry fields using air conduction tests

(Convery et al., 2014). In this sense, there is a

systematic review of works that check the validity of

automated threshold audiometry compared with the

gold standard of manual threshold audiometry

(Mahomed et al., 2013). Therefore, there is a need to

allow human experts to participate in this process by

providing them with several audiometry data sources

and enabling their analysis. The current work

presents a framework able to process audiometer

images in order to extract information which can be

useful to analyse subject's hearing levels.

The remainder of the paper is structured in the

following sections. The second section depicts the

audiometry context in which the proposed

framework has been developed and tested. This

framework is outlined in section 3 and the fourth

section reports the obtained Results. Section 5

describes some related works Finally, some

Conclusions and further works are drawn.

2 AUDIOMETRY CONTEXT

Audiometry can be considered as a tool to measure

the subject’s hearing capability according to

different sound frequencies. There are several

methods to measure this capability and they can be

divided into subjective and objective audiometry.

Mendel (2007) emphasized the need for both

subjective and objective documentation of hearing

aid outcomes. In this case, the current work focuses

on subjective measures as a way to get audiometry

information by means of specific hearing tests. Pure

tone audiometry (PTA) is measured in dB HL

(Hearing Level) and this value is used to identify the

hearing threshold level of an individual. This level

represents the higher intensity of sound to be

perceived by a subject, compared with people who

have a normal hearing level.

For this work, a modified audiometer called

TLTS (Tomatis, 2016) has been used, which is

based on the use of de SPL (Sound Pressure Level)

values as the difference between the pressure

produced by a sound wave and the barometric

pressure. TLTS was designed by Dr. Alfred Tomatis

using a curve of absolute hearing threshold values

and it is used for performing a specific listening test

that registers hearing levels once these are almost

inaudible. The listening test evaluates an

individual’s auditory thresholds in terms of

frequency, ability to identify the source of sounds,

ability to discriminate between frequencies, and

auditory laterality. The analysis of the resulting

curves serves to determine the person’s quality of

listening and from this to induce a psychological

profile. This kind of tests has been performed by

professionals of the Isora Solutions company who

are participating in a research project about the

effect of neurosensory stimulation to improve

listening skills (Perez et al., 2016). There are

multiple types of actions which can be performed to

determine subject's hearing levels in this context.

Next subsections describe such actions and the

obtained outcomes to be further processed.

2.1 Audiometric Tests

Four main types of audiometric behavioural tests

have been performed which address different

hearing parameters:

 Thresholds

 Laterality

 Selectivity

 Availability

Threshold of hearing is the minimum sound of level

that a human ear can perceive in a certain frequency

band and it is considered as a measure of hearing

sensitivity. This kind of sensitivity can be

represented using a chart called audiogram that

displays the audible threshold intensity for

standardized frequencies. Figure 1 shows an

example of audiogram that represents intensity

thresholds measured using dB SPL values (displayed

on the vertical axis), which change as frequency

ranges from 250 to 8000 Hz (horizontal axis). In this

audiogram, blue lines are associated to the air

conduction while red line symbols refer to the bone

conduction and the green line to the availability.

Both via air and via bone conduction (using a

vibrator placed on the top of the head) are the main

data sources in the TLTS tests. It is important to

remark the difference in the sound speed of the two

mediums since the travelling time of the bone

conduction to the brain is assumed to be faster that

the air conduction. According to Dr. Tomatis, the

bone conducted sound serves as a wakeup call to

prepare the brain for incoming sound. Then, the

delay between bone and air-conducted sound has to

be measured.

SIGMAP 2017 - 14th International Conference on Signal Processing and Multimedia Applications

Figure 1: Sample of TLTS audiogram to display Threshold curves.

The second test is based on checking laterality,

which is only obtained from a TLTS device, as a

measure to observe how humans focus their hearing

on one ear (left or right). This measure uses two

main values called Extraction (Ext.) and Resistance

(Res.) representing the laterality profile. Figure 2

shows a table that allows the matching between

these two values and the laterality levels.

Figure 2: Laterality TLTS dataset.

Selectivity refers to the ability to differentiate the

pitch of sounds in relation to each other, but also the

direction of variation in pitch. The selectivity test

determines the maximum opening of the subject’s

auditory that is obtained by the frequencies in which

some kind of barrier is detected.

In this case, selectivity is obtained from 35

frequency options. Each of these values can be

marked when a hearing misunderstanding is

detected. An overall percentage of marks can be

computed from this test. The higher is this

percentage, the lower is the subject’s level of

listening and memory abilities. Figure 3 shows a

sample of this test where a barrier mark is observed

for the left ear in the 3250 Hz frequency,

Figure 3: Selectivity test sample.

A Computer-based Framework to Process Audiometric Signals using the Tomatis Listening Test System

These tests were implemented with the help of

the TLTS device, such as the one displayed on

Figure 4. This type of device determines hearing

levels, but also measures the ability to discriminate

between different sound intensities. The TLTS offers

built-in wavefiles for a variety of speech extended

high frequency evaluation and PTA calculations. It

provides features to process speech information

through live voice, mp3 recordings and wavefiles as

well as word recognition capabilities. For tone

analysis, several air conduction and bone condition

mechanisms are addressed, and the possibility to

manage masking information.

Figure 4: Picture of a TLTS device.

2.2 Test Outcomes

This subsection describes some of the outcomes

obtained in the TLTS process. Such outcomes are

divided in two categories: spreadsheets files and

image files. Intensity thresholds and availability are

represented by means of audiogram charts such as

the one displayed on Figure 1. The upper area shows

two polygonal lines that represent those threshold

diagrams associated to the air conduction (blue line)

and bone conduction (red line) for the right ear in

this case. The lower area shows a set of characters

that represent special situations in a hearing scenario

Laterality and selectivity data are stored on

spreadsheets files in a tabular format such those

displayed on Figure 5. The first table shows part of

the values returned by the audiometer that compute

selective frequencies for the left ear either in air

conduction test (LAC) or bone conduction (LBC),

and, similarly, for the right ear (labelled as RAC and

RBC, respectively).

Figure 5: Selectivity and Laterality data tabular display.

3 FRAMEWORK DESCRIPTION

The framework proposed in this work deals with

processing the different signal sources mentioned in

previous sections. Figure 6 shows an overview of the

framework functionality. It is structured in two main

flows: the first one addressed to process the

audiogram images and the second one in charge of

selecting spreadsheet data. The results of both flows

are integrated in order to analyse and interpret them.

Figure 6: TLTS Framework overview.

SIGMAP 2017 - 14th International Conference on Signal Processing and Multimedia Applications

3.1 TLTS Processing

TLTS processing is based on the use of OpenCV

(Open Source Computer Vision Library). OpenCV is

an open source computer vision and machine

learning software library that was built to provide a

common infrastructure for computer vision

applications. It has C++, C, Python, Java and

MATLAB interfaces and works on multiple

operating systems. This library provides several

features that help to gather quantitative information

from different kinds of images. In the current case,

images come from audiograms such as the one

displayed on Figure 1 and the use of OpenCV

enables to get the locations of points that compose

the audiometric lines.

These locations assign values of intensity

thresholds (measured in dB) for each frequency

value in the horizontal axis. Original audiogram

images are bitmaps of resolution 1200x850 pixels. A

first processing step consists in identifying each of

the lines by colour, and segmenting it from the rest

of the image. Figure 7 shows the red line that

represents the threshold bone conduction line in the

target audiogram.

Figure 7: Segmented TLTS image.

The second step is about computing the

measured values of each point of a segmented line.

Checked points for each of the values of the

frequencies can be located from the analysis of the

labels in the horizontal axis, providing a frequency

value in Hz. By projecting these point locations in

the vertical axis, the dB value can be computed. In

order to find such points a template matching

process is performed do detect the point shapes. The

coordinates of the center of mass of these shapes are

returned using an iterative sequence over the

audiogram colour lines. Figure 8 shows a part of the

audiogram in which a green line is marking the

detected points.

Figure 8: Point shape detection.

Figure 9: Segmentation of sound TLTS spatialization.

A similar matching procedure is used to filter the

spatialization symbols as they are displayed on

Figure 9, in which “?” characters are marked within

red squares. At the end of the image processing,

tabular formats with the collected data are produced

and converted to text files (e.g. csv documents).

3.2 Spreadsheet Processing

The process of gathering information from

spreadsheets is based on the use of scripts which

allow researchers to get a web view of audiometric

data.

Figure 10 shows an example of screenshot that

displays Laterality values from a selected dataset.

A Computer-based Framework to Process Audiometric Signals using the Tomatis Listening Test System

Figure 10: Script outcomes for selecting laterality data items.

These scripts provide access to spreadsheet files

either individually or gathering a set of them, in

order to get and retrieve the relevant data items and

store such items in tabular formats ready to be

analysed in further steps.

4 RESULTS

Results obtained by the framework components are

evaluated in this section. With this aim, several

experiments have been implemented that check the

framework accuracy and adequacy for processing

audiometry TLTS signals. These samples represent

different measures of audiometric tests that were

available in a tabular format after the framework

application.

A first evaluation consisted in comparing values

obtained from processing audiograms with original

tabular values. Table 1 shows a list of threshold

values coming from audio conduction tests for a

specific subject whose audiograms were processed

using the framework. These values are structured

into several columns that display the audiogram

frequency, the threshold data for the Left ear on Air

Conduction (LAC), or Bone Conduction (LBC), the

extracted points (Ext.) from both graph audiograms,

and the Average difference between these data.

Table 1: List of threshold values (measured and extracted).

Freq.

LAC

Ext.

LBC

Ext.

Avg.Dif..

125

54,915

0,0425

250

39,783

49,953

0,132

500

49,953

59,876

0,0855

625

30,109

44,744

0,0735

750

29,86

44,744

0,198

875

34,822

39,783

0,1975

1000

34,822

39,783

0,1975

1125

14,729

34,822

0,2245

1250

14,729

19,69

0,2905

1375

14,729

19,69

0,2905

1500

9,767

24,899

0,167

1625

14,729

19,69

0,2905

1750

14,729

0,271

1875

19,69

24,899

0,2055

2000

14,729

24,899

0,186

2250

-0,155

19,69

0,2325

2500

4,806

14,729

0,2325

2750

9,767

14,729

0,252

3000

14,729

0,271

3500

14,729

0,271

4000

19,69

14,729

0,2905

4500

24,899

19,69

0,2055

5000

19,69

29,86

0,225

6000

34,822

29,86

0,159

8000

34,822

29,86

0,159

SIGMAP 2017 - 14th International Conference on Signal Processing and Multimedia Applications

Figure 11: Audiogram processed sample with detection of threshold curves.

Such list of values is based on the audiogram

sample displayed on Figure 11 and the average

difference is less than 0.3 in every processed data

item either for air or bone conduction. These

threshold data items can be further processed to

analyse their correlation using a scatter chart such

the one displayed on Figure 11 that displays the

relationship between air and bone sound conduction.

Figure 12: Correlation of TLTS threshold curves.

The results of processing spreadsheets files are

also addressed in the framework evaluation. These

files store information about audiometric parameters

such as Laterality and Selectivity issues together

with additional user information (e.g. Age). An

example of Laterality profile regarding a specific

subject is displayed on Figure 10. The framework

can provide a collection of different subject profiles

in order to allow their analysis. Figure 12 shows a

radar chart that displays the relationship between the

values of Resistance and Extraction as Laterality

parameters and the age of nine subjects in a study

sample.

Figure 13: Analysis of TLTS laterality information.

A Computer-based Framework to Process Audiometric Signals using the Tomatis Listening Test System

Figure 14: Script outcomes displaying the collection of selectivity data items.

Figure 15: Selectivity bar chart to display threshold distribution.

SIGMAP 2017 - 14th International Conference on Signal Processing and Multimedia Applications

selectivity information can also be analysed using

the spreadsheets processed in the framework.

For example, Figure 14 shows the display of

frequencies that were marked in a specific

Selectivity case and

Figur represents the bar chart for that case in

which frequencies associated to air sound

conduction in the left ear (LAC) and the right ear

(RAC) are compared with those produced via bone

sound conduction (LBC and RBC, respectively).

5 RELATED WORKS

Computer-based frameworks for managing and

processing audiometry signals are rather unusual as

long as these signals have a strong link with

audiometers or specific devices that generate and

produce them. Some organizations such as NCHAM

at Utah University or HIMSA

have implemented

software to provide audiology professionals with

systems to manage patient information. In the case

of NCHAM, they provide HiTrack

as an Early

Hearing Detection and Intervention (EHDI) Data

Management System to screen and register relevant

hearing information.

One of the HiTrack main advantages is the

possibility to connect with screening equipment

coming from multiple manufacturers. HIMSA has

developed NOAH as an integral framework allowing

“hearing instrument fitting, audiologic measurement

and database management system”. This software is

able to store patients’ audiologic profiles from

different suppliers and also record notes regarding

patient sessions.

Additionally, the Audiology and Speech

Pathology Software Development Group at

Memphis State University developed a program for

the Abbreviated Profile of Hearing Aid Benefit (Cox

and Alexander, 1995) to document the outcome of a

hearing aid fitting and to compare and evaluate the

same fitting over time.

Otherwise, manuals and audiometry guides can

be found (NCHS, 2016) but they are usually limited

to establish procedures and recommendations in the

use of instrumentation or how to record audiometry

results. There are authors such as Dau (2008), and

Jepsen et al., (2008) who have proposed models for

auditory processing and, in the last case, their

authors include a Computational Auditory Signal

Processing (CASP) framework that was

implemented in Matlab / Octave scripting language.

https://www.himsa.com/

http://www.hitrack.org

However, these models are usually focused on

specific aspects like masking in human listeners. In

this sense, Heinz (2010) also presented a

computational model of sensorineural hearing loss.

This work focused in the context of a sensorineural

stimulation project with the aim to process a wide

spectrum of audiometry signals beyond the typical

hearing loss situations. Therefore, the priority was to

introduce a framework that could be adapted to

process alternative source of data, either in tabular

format or using image files, and provide audiology

researchers with computer tools to analyse them.

6 CONCLUSIONS

The current work has presented a computer-based

framework that deals with the processing of

audiometry signals coming from different data

sources. These audiometry signals have been based

on audiogram images and spreadsheet files produced

by the TLTS device that were addressed to specific

hearing tests. Obtained outcomes show the wide

range of possibilities of the proposed framework to

compute these data sources and contribute to

improve the assessment of hearing tests using them.

For example, by determining the accuracy in the

processing of audiogram charts whose data can be

used for statistical studies. Another framework

contribution is its potential to retrieve multiple data

sources and combine them to produce graphic charts

that allow audiologists to easily envision hearing test

outcomes. This feature remarks the framework

adaptability to fit processing functions according to

the required needs. Further works plan to

incorporate new processing procedures and integrate

them in a Web portal that enables a universal access

to the framework services.

ACKNOWLEDGEMENTS

Thanks to the support of the Research Project

“Neurosensorial Stimulation for the Integration of

English Language” Universitat Politècnica de

València & Isora Solutions, 2016, the Computer

Engineering department (DISCA) and the ETSINF

(Escola Tècnica Superior d’Enginyeria Informàtica)

at the Universitat Politècnica de València.

A Computer-based Framework to Process Audiometric Signals using the Tomatis Listening Test System

REFERENCES

Boothroyd A., 2006. CasperSent: an example of computer-

assisted speech perception testing and training at the

sentence level, Phd thesis.

Convery E., Keidser G., Seeto M., Freeston K., Zhou D.,

Dillon H.,2014. Identification of conductive hearing

loss using air conduction tests alone: Reliability and

validity of an automatic test battery. Ear and Hearing

35(1): 1–8.

Cox R M, Alexander G C. The abbreviated profile of

hearing aid benefit. Ear Hear 1995; 16: 176–186.

Dau, T., 2008. Auditory processing models. In D.

Havelock, S. Kuwano, & M. Vorländer (Eds.),

Handbook of signal processing in acoustics, Vol. 1

(pp. 175–196). New York: Springer.

Eisenberg L.S., Martinez A.S., Boothroyd A., 2007.

Assessing auditory capabilities in young children. Int.

Journal Pediatr Otorhinolaryngol; 71: 1339–1350.

Fernández A., Ortega M., Penedo M.G., 2014. Computer

Aided Hearing Assessment: Detection of Eye Gesture

Reactions as a Response to the Sound. In: Campilho

A., Kamel M. (eds) Image Analysis and Recognition.

ICIAR 2014. Lecture Notes in Computer Science, vol

8815. Springer, Cham.

Heinz M.G., 2010. Computational modelling of

sensorineural hearing loss, Computational models of

the auditory system.eds. Meddis R, Lopez-Poveda EA,

Popper AN, Fay RR (Springer, New York), pp. 177–

202.

Jepsen M.L, Ewert S.D, Dau T., 2008. A computational

model of human auditory signal processing and

perception. J. Acoust Soc AM in press, 124(1).

MacKersie C.L., Boothroyd, A., Minnear. D., 2001.

Evaluation of the Computer-Assisted Speech

Perception Test (CASPA) Journal Am. Acad. Audiol.,

pp. 390–396.

Mahomed F., Swanepoel D.W., Eikelboom R.H. & Soer S.

2013. Validity of automated threshold audiometry: A

systematic review and meta-analysis. Ear Hear, 34,

745–752.

Mendel, L.L., 2007. Objective and subjective hearing aid

assessment outcomes. American Journal of Audiology,

16, 118-129.

NCHS, 2016. Audiometry Procedures Manual. National

Center for Health Statistics. https://wwwn.cdc.gov

/nchs/data/nhanes/2015-2016/manuals/2016_

Audiometry_Procedures_Manual.pdf

Perez, C., Garcia, C., Conejero, M., Capitán, A., Cerna,

H., 2016. Neurosensorial Stimulation for the

Integration of English Language. http://www.tomatis.

com/es/el-metodo-tomatis/actualidad/neurosensorial-

stimulation-for-the-integration-of-english-language-

en.html

Tomatis, 2016. User manual (Level II. TLTS).

www.centroauditivo-valencia.es.

SIGMAP 2017 - 14th International Conference on Signal Processing and Multimedia Applications