1 / 19

Building a corpus of pathological speech

Building a corpus of pathological speech. Gwen Van Nuffelen Marc De Bodt. Catherine Middag Jean-Pierre Martens. Dutch Corpus of Pathological and Normal Speech. disturbed muscular control due to damage of the nervous system  weak, slow, imprecise, uncoordinated movements.

hung
Download Presentation

Building a corpus of pathological speech

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Building a corpus of pathological speech Gwen Van Nuffelen Marc De Bodt Catherine Middag Jean-Pierre Martens

  2. Dutch Corpus of Pathological and Normal Speech disturbed muscular control due to damage of the nervous system  weak, slow, imprecise, uncoordinated movements

  3. Dutch Corpus of Pathological and Normal Speech TL: surgical removal of the larynx and separation of the trachea from the mouth, nose, and esophagus TE, E, electro larynx (servox) PL: partial removal of laryngeal structures, vocal folds

  4. Dutch Corpus of Pathological and Normal Speech

  5. Speakers • native speakers of Dutch • adequate language, cognitive, visual and hearing* abilities

  6. Recordings • Natural, quiet environment ~ clinical setting • No sound treated box • Mini-disc (Sony, MZ-R700) • Microphone • Sony (mouth-microphone distance: 30 cm) • Shure head set • Transferred to a notebook  wave file (mono, 44kHz) • 16 kHz

  7. Type of samples

  8. Dutch Intelligibility Assessment (DIA) • Intelligibility at phoneme level • 50 consonant – vowel – consonant words • 3 subtests: • A: initial consonants (19 words) • B: final consonants (15 words) • C: medial vowels/ diphthongs (16 words) • Balanced mix of existing and non-existing (well pronounceable) words • Large pool of test items: 25 lists/ subtest  25*25*25 different tests

  9. DIA 16 year-oldgirl, stroke, dysarthria, PI: 40% 79 year-old male, TL, TE-speech, PI: 68%

  10. top DIA List A10 Intelligibility: percentage of phonemes correctly understood

  11. DIA

  12. Annotations DIA • Praat • 2 tiers • Tier 1: target word • Tier 2: fixed frame + perceived phoneme • . VC • CV. • C.C • Orthographic transcriptions

  13. List A Target phoneme: initial consonant Fixed frame: . V C

  14. Articulation assessment • Children • Insufficient reading skills • Logo-Art (Baarda et al, 2001) • Picture naming test • Annotations: • Orthographic • Tier 1: target • Tier 2: perceived utterance (no fixed frame)

  15. Sentences • Motor Speech Profile (Kay Elemetrics) • ‘Wil je liever de thee of de borrel ?’ • ‘Na nieuwjaar was hij weeral hier’ • N= 211 • Orthographic transcriptions • Tier 1 – tier 2; no word boundaries man, no speech pathology 18 year-old male, congenitaldysarthria

  16. Text Marloes and Text • Text ‘Papa en Marloes’ • standardized text • balanced representation of Dutch phonemes • often used in clinical practice • Text • different texts with the same reading level • orthographic transcriptions • 2 tiers • boundaries between sentences

  17. (Semi) Spontaneous speech • Spontaneous • Semi spontaneous: randomly selected sequence of pictures • No annotations available

  18. Future • Gradually increase number samples • DIA  validation SPACE intelligibility assessment • DIA sentence level: > 200 control speakers 3*6 sentences + annotations + pathological samples

  19. Thank you!

More Related