by Buddha NeeharikaDAIICT and Prof'Hemant Patil DAIICT

1
by Buddha Neeharika(DA-IICT) and
Prof.Hemant Patil (DA-IICT)

• O-COCOSDA 2007,
  HANOI
• CORPORA FOR INFANT CRY ANALYSIS
  December 4th, 2007

Place photo here
2
OVERVIEW

• Introduction
• Recent works in the this field
• Challenges associated
• Corpus details
• Experiences during data collection
• Empirical observations
• Observations from spectrogram
• Problems to be addressed

3
INTRODUCTION

• Cry is the primary means of communication for
  infants.
• Infant cry a possible tool for early diagnosis
  
  
• 5
• Neurolinguistics considers the just born baby cry
  as the first linguistic manifestation.

4
First Cry

• It is during the first cry that ones lungs get
  opened for the first time
• Really crutial for ones life!!
• This cry is one of the five parameters (Apgar
  count) to assess the childs condition soon after
  birth
• Parameters of apgar count
• 1.Strength and regularity of heart rate
• 2.Lung maturity
• 3.Muscle tone and movement
• 4.Skin color / oxygenation
• 5.Reflex response to irritable stimuli --gtcry
• In a normal new born child the first cry is
  expected to be a vigorous one. A varied cry is
  also a way of interpreting the condition of the
  child

5
Normal cry for comparision.

• Normal cry for 15 sec

• First cry for 15 sec

6
Challenges associated

• Data collection not as simple as it is for grown
  up speech.
• Organising the data.
• The pitch frequency order goes as follows
• InfantsgtchildrengtFemale(Adult)gtMale(Adult).
• Since, the frequency of the infants sound is very
  high,
• the problems associated with appropriate sampling
  and aliasing come into picture.
• This makes this task highly challenging.

7
Recent works in this field

• . Till date, most of the works are done with
  respect to trying to distinguish the normal cries
  from abnormal cries and further variation amongst
  the diseases such as Hypoxic ischemic
  encephalopathy ,respiratory disorders,etc based
  on pproaches such as Adaptive neuro-fuzzy
  inference systems(ANFIS)1,,Levels-of-distress(LO
  D) analysis2
• There is a study done on the evolution of the
  fundamental frequencies for the first seven days
  of 172 infants.3

8
CORPUS DETAILS

• Cries of 200 infants of various backgrounds were
• recorded at
• 1).King George Hospital(K.G.H),
  Visakhaptnam(India)
• 2).Prabha Nursing Home(PNH), Visakhapatnam
  (India)
• 3).Child Clinic, Visakhapatnam (India)

9
Presence of background noise
10
Environmental conditions
11
Technical recording conditions

• Digital Recorder
• Due to the inherent limitations of collecting
  data from many infants, a portable Cenix digital
  voice recorder was used for this purpose.
• Specifications ModelVR-P2340,
• Memory 64MB,
• Storage media Imbedded flash memory ,
• Sampling Rate-12KHz,Sampling type PCM 16bits.
• Actual maximum output 150 mW
• Microphone
• The microphone was held approximately 6-10cm away
  from the infants mouth, so as to avoid any
  clipping of the data.

12
Meta Data

• Details of the infants such as
• Name, age, weight.ailments (if any) and the
  corresponding system that is affected, whether
  the child is being brought up by the parents
  themselves are noted.
• Some comments regarding the disease were also
  documented.
• Besides, wherever possible , the commentary of
  the physician or doctor who diagonised the infant
  was recorded.
• The documented meta data and commentary (if any)
  corresponding to the infants is stored in .txt
  file and .wav files respectively, in the
  corresponding folders.

13
CORPUS DESIGN

• Based on the inputs from the doctors and the
  parents/ guardians, the cries are classified as
  follows
• Normal Infant cries
• Normal infants with abnormal cries.
• Extremely sick new borns
• Pre-term born cries
• New born first cries

14
Experiences during data collection(1/2)

• Initially, the parents or the guardians of the
  infants were informed of the purpose of recording
  and their signatures regarding their consent for
  the same was taken.Fortunately, in all the cases,
  the presence and consent of the concerned doctor
  was crucial .Otherwise, the parents would not
  have understood the purpose of the study
• Almost 60 of the recordings were spontaneous
  cries. Patient monitoring of the infants was
  required for the same.
• In the hospitals, the situation was such that
  while recording for one infant, another starts
  crying this required swift movements from infant
  to infant.

15
Experiences during data collection(2/2)

• owing to the fact that infants cry mostly
  throughout nights and in the morning till around
  10 am, recordings in the hospitals could be done
  only in the early hours of the days. Hence, there
  was a sort of virtual time constraint after which
  the infants would sleep.
• One new borns cry was recorded for 16 days (on
  alternative days) during the first one month of
  his life in the home environment. This was done
  to capture the cries of a normal infant at night
  time also .However, due to certain constraints
  this could not be done for other infants.
• Recording the pre-term baby cries was really
  difficult, because they hardly cry. Whenever they
  cry their voice is so shrill that it would be
  hardly audible. So these infants required keen
  attention
• Recording the first cry, of the new born was a
  different experience all-together. As soon as the
  child was delivered, it was handed over to the
  pediatrician for general health check up. At this
  time when the new born was placed in the
  incubator, the recordings were done.

16
EMPIRICAL OBSERVATIONS

• In general infant cries come under voiced and
  unvoiced sound category , because sounds are
  produced by forcing air through the glottis with
  the tension of the vocal cords adjusted so that
  the vocal folds tense up and are brought close
  together, thereby producing quasi-periodic pulses
  of air which excite the vocal tract
• It has been observed that the cry of a just new
  born infant and days old infant varies
  significantly
• There is an interesting connection between the
  amplitude of cry and weight of the baby cry of
  low birth weight infants is generally shrill, as
  compared to normal healthy babies .
• An interesting observation is that infants
  generally cry throughout the nights till around
  1000 am in the morning. Scientifically, reason
  for this is that this sleeping pattern is
  synchronous with their sleeping pattern in
  mothers womb. So most of the recording work in
  hospitals had to be done in the early hours of
  the day

17
Features of the corpus

• Distribution of the samples over gender
• Distribution of the samples over age.

18
NOTE

• Erratum
• All the formant frequncies specified in the table
  of spectrogram analysis are in Hz.
• They are specified in Khz by mistake.
• Kindly consider the same in the paper.

19
OBSERVATIONS BASED ON SPECTROGRAMS
20
HIE
21
Respiratory Disorder
22
First Cry
23
PROBLEMS TO BE ADDRESSED

• The following issues will be dealt with
  appropriate techniques of pattern recognition and
  neural networks
• (i)Classifying several diseases
• (ii)Analysis of the first cry in lines of
  apgar count.
• (iii)Analysis of pre-term baby cry.
• (iv) Emotion extraction from the cry.

24
References for those quoted in the presentation

• 1 Orion F.Reyes-Galaviz, Emilo Arch Tiarado and
  Carlos Alberto Reyes-Garcia Classification of
  infant crying to identify pathologies in recently
  born baies with ANIFS ICCHP 2004, LNCS 3118,
  pp. 408-415, 2004.
• 2 Qiaobing Xie,Rabab K. Ward , and Charles A.
  Laszlo Automatic assessment of Infant's
  Level-of-Distress from the Cry signals IEEE
  Transactions on speech and audio Processing , Vol
  4, No.4, July 1996
• 3Katarina Michelson, Kenneth Eklund, Paavo
  Leppanen,HEikki Lyytinen,Cry Characteristics of
  172 healthy infants 1-to-7 day-old infants,
  Folla Phoniatrica 200254190-200
• 4 A. I. Tsyplikhin Analysis of Vocal Pulses
  in a Speech Signal, ISSN 1063-7710, Acoustical
  Physics, 2007, Vol. 53, No. 1, pp. 105118.
• 5 http//www.hindu.com/seta/2003/10/02/stories/2
  003100200330300.htm

25
Thank You!!!

• Thank you!!