Docstoc

626

Document Sample
626 Powered By Docstoc
					http://despho-apady.univ-avignon.fr/




    The DesPho-APaDy Project:
   Developing an Acoustic-phonetic Characterization
            of Dysarthric Speech in French

 C. Fougeron*, L. Crevier-Buchman*, C. Fredouille°, A. Ghio^, C. Meunier^,
 C. Chevrie-Muller, J.-F. Bonastre°, A. Colazo-Simon, C. Delooze, D. Duez^,
     C. Gendrot*, T. Legou^, N. Lévêque*, C. Pillot-Loiseau*, S. Pinto^,
      G. Pouchoulin°, D. Robert^, J. Vaissiere*, F. Viallet^, C. Vincent*

         * Lab. de Phonétique et Phonologie, UMR 7018 CNRS-Paris3/Sorbonne Nouvelle, Paris
                                ° Université d'Avignon, CERI/LIA, Avignon
            ^ Lab. Parole et Langage, UMR 6057 CNRS Aix-Marseille Univ., Aix-en-Provence
                    What is Dysarthria?
• Neurologically-based speech disturbances resulting
  from damage to the central and/or peripheral nervous
  system that impair the transmission of neural messages
  to the muscles involved in speech production
     – Deficit in the motor execution at any level of speech production:
             • respiratory
             • phonatory
             • supralaryngeal (velopharyngeal and orofacial)

     – Requires the consideration of multiple speech parameters




2010-05-20            Fougeron et al.: The DesPho-APaDy Project, LREC 2010   2
Rationale and Objectives of the Project

- To constitute a searchable speech database of French
   dysarthric patients:
      1. for the preservation of existing databases
      2. for the need of the acoustic description

- To provide a systematic, quantified acoustic
description of the speech patterns of French
dysarthric speakers



2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   3
The Multiple-Field Query Database:
                                 Why?

• Storage and organization of digitized dysarthric speech
  corpora (acoustic and physiological data) and associated
  patients’ clinical information

• Accessibility (for selected and protected use) through
  multiple-field queries (M/F, age, on/off stimulation,…)




2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   4
The Multiple-Field Query Database:
                          Which data?

 The CCM (Claude Chevrie-
  Muller) Corpus recorded over the
    last 30 years,
  ~1000 hours of disordered speech,
    5000 patients – adults and
    children
 * 860 patients classified according
    to their neurological diagnosis
 * 60 control speakers

 Data recorded
 Sound, EGG (ElectroGlottoGraph)
2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   5
The Multiple-Field Query Database:
                          Which data?
• The ANH (Aix-Neurology-
  Hospital)
 Corpus recorded for the past 15
  years:
  * 990 patients,
  * 160 control speakers

Data recorded
Sound, aerodynamic.
                                                         EVA2


2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   6
The Multiple-Field Query Database:
                                        How?
• Getting the audio files:
     – Digitization of the analog 1/4” audio tapes
             • speed change during each recording session
             • “real time” auditory control of the recording
             • Problems encountered: poor quality recordings, some tapes
               in bad condition
     – Transfer of DAT tapes
             • To date, data from 180 patients have been digitized (analog
               tapes), 94 have been captured from DAT
     – Segmentation per patient and per speech task (idem
       for the control population)
     – Data were renamed for anonymous storage

2010-05-20          Fougeron et al.: The DesPho-APaDy Project, LREC 2010     7
The Multiple-Field Query Database:
                                   How?
• Designing the database:
     – Choice of a relational model to structure the data
     – Definition of a set of objects, their attributes and the
       relation between them, related to the clinical
       environment
     – Refinement by confronting the relational data model
       with empirical and “real” clinical data
• Developing the query interface:
     – Basic sociolinguistic information
     – Clinical information
     – Recording session information

2010-05-20     Fougeron et al.: The DesPho-APaDy Project, LREC 2010   8
        Dysarthric Speech in French:
Dimensions to be considered when dealing with dysarthria


• Various parameters can be affected…
      • phonation
      • articulation
      • timing/rhythm
      • prosody
      • fluence…

• Need of appropriate and sufficiently large quantities of
  speech data, homogenous in quality, and sufficiently
  documented by clinical information on the patient


2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   9
         Dysarthric Speech in French
                               Methods
• Methods and analysis procedures based on both
  phonetics and speech engineering approaches
   – Manual analysis (by human experts) of the acoustic
     properties of the productions
   – Automatic acoustic analysis of speech signals
   – Continuous back and forth between these two
     techniques should gain from the potential of both
     approaches




2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   10
      Dysarthric Speech in French:
  Selection of Patients (Acoustic Study)
• Focus on neurophysiologic alterations of 3
  neurologic systems:
     – Pyramidal system: ALS dysarthria (30 patients)
     – Cerebellar system: ataxic dysarthria (30 patients)
     – Extrapyramidal system: Parkinsonian dysarthria (30
       patients from the ANH corpus)
• Selection based on:
     – The clinical information, (the certainty of the
       diagnosis, the ongoing treatment, patient
       demographics)
     – The severity of the dysarthria (work on dysarthria with
       relatively intelligible speech)
2010-05-20    Fougeron et al.: The DesPho-APaDy Project, LREC 2010   11
        Dysarthric Speech in French:
      Pre-Processing of the Audio Files
• In order to be able to perform both the
  manual and automatic acoustic analyses,
   Need of:
     – Orthographic transcriptions



     – Automatic text-constrained phonetic
       alignment

2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   12
      Pre-Processing of the Audio Files:
              Orthographic Transcriptions
                  Automatic Alignment
     1. Orthographic transcriptions
     <s> [ext] *vu* *puve* *i* *ale* [ext]*alOR* Mr
       Seguin n'avait jamais eu de bonheur avec ses
Each speech production which differs from the original text
  was annotated using the SAMPA phonetic alphabet
   2. Automatic text-constrained phonetic alignment




2010-05-20   Fougeron et al.: The DesPho-APaDy Project, LREC 2010   13
             Dysarthric Speech in French:
             Automatic Phonetic Alignment
• Evaluation of the automatic phonetic alignment
  comparing the production of phonemic labels and the
  boundaries annotated by two phoneticians

                                            Male               Female

                    control                   X                    X

              Moderate dysarthria             X                    X

               Severe dysarthria              X                    X




2010-05-20        Fougeron et al.: The DesPho-APaDy Project, LREC 2010   14
             Dysarthric Speech in French:
             Automatic Phonetic Alignment
Results:
When comparing segment time boundaries
  produced manually and automatically, a shift
  above 20ms is observed:
      17% of control speakers,
         24% of moderate dysarthric patients
           56% of severe dysarthric patients
But strong agreement (manual/automatic) for
  spectral measurements
2010-05-20      Fougeron et al.: The DesPho-APaDy Project, LREC 2010   15
             Conclusions/Perspectives

Encouraging results for clinical research on speech
  disorders and speech science.

Observing type and range of variations due to motor control
  deficit can help for a comprehensive model of speech
  variation for normal speakers.

A better understanding of the variations that permit to
  consider a dysarthric speech as deviant could provide
  insights to the boundary between normal and
  pathological speech.

2010-05-20    Fougeron et al.: The DesPho-APaDy Project, LREC 2010   16
                  Acknowledgments

Funded by the ANR BLAN08-0125 of the
 French National Research Agency

We deeply thank P. Clément, A. Nurenberg
 and O. Panseri who are also collaborating
 to this project

         For further information, please visit:
         http://despho-apady.univ-avignon.fr
2010-05-20    Fougeron et al.: The DesPho-APaDy Project, LREC 2010   17

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:6
posted:1/23/2011
language:English
pages:17