Acoustic analysis of occlusive weakening in Parkinsonian French speech Danielle Duez To cite this version: Danielle Duez. Acoustic analysis of occlusive weakening in Parkinsonian French speech. International Congress of Phonetic Sciences, Aug 2007, Saarbrücken, Germany. pp.1-4. hal-00244491 HAL Id: hal-00244491 https://hal.archives-ouvertes.fr/hal-00244491 Submitted on 7 Feb 2008 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. ACOUSTIC ANALYSIS OF OCCLUSIVE WEAKENING IN PARKINSONIAN FRENCH SPEECH Danielle Duez Laboratoire Parole et Langage, CNRS UMR 6057, Université de Provence, France [email protected] ABSTRACT Occlusive weakening is a complex process which may affect languages differently. For example, The current study investigated selected acoustic French and English are known to differ when they characteristics of the weakening of occlusives instantiate the voicing contrast: voicing-related (O’s) in French Parkinsonian Speech (PS). The durations are larger in French than in English and results confirm an increase in the reduction of O’s /b, d and g/ closures are more consistently in PS compared to control speech (CS). In PS, O’s phonated in French than in English. The effect of have a significantly decreased intervocalic energy context is also language-specific: e.g. in French, level, slightly shorter realisations and a higher there is a strong tendency to nasalisation in the number of visible formants and noise; the number vicinity of nasal vowels. of burstless and omitted O’s is also higher. The current study analysed selected acoustic However, weakening patterns vary between characteristics of O-weakening in French PS. It different consonants and are strongly dependent was anticipated that the information obtained on on voicing and place. Occlusive weakening, a O’s-acoustic patterns in read PS might increase consequence of Parksinson’s disease production our comprehension of whether and how PD deficits, appears to be influenced by the inherent- impairs the acoustic patterns of speech sounds. articulatory characteristics of consonants. It was assumed that O-weakening results in greater reduction and assimilation to context than 1. INTRODUCTION in normal speech. Reduction means an obscuration process which results in a loss of A wide range of studies have examined the impact features or an increase in sonority (voicing, of Parkinson’s disease (PD) on the production of fricativisation, and sonorisation). Assimilation occlusives (O’s); all observed significant refers to a process which increases the similarity weakening reflected at the acoustic level by between two adjacent (or next-to-adjacent) shorter durations and alterations in spectral segments. To achieve this, acoustic patterns of patterns. For example, voiceless O’s, which are O’s in PS and control speech (CS) were compared normally associated with a silent gap, tend to in a standard text read by nine French PD patients exhibit energy during the silent gap [1, 2 and 3]. and eight healthy control subjects. O’s acoustic This energy is typically expressed as either analysis was made as a function of voicing, place turbulence noise (spirantization), generated at the and vocalic context (oral and nasal). site of oral constriction because of an incomplete occlusion, or voicing energy which occurs as a 2. METHOD result of poor coordination between laryngeal and supralaryngeal gestures [3]. Data obtained for 2.1. Subjects voice onset time (VOT), defined as the acoustic interval between the burst and vowel onset also The data for this study were collected from 17 indicate laryngeal-supralaryngeal discoordination. French native male speakers composed of 9 The VOT values for voiced and voiceless O’s individuals diagnosed with PD and 8 age and form non-overlapping distributions with a gender matched control speakers. The PD boundary at about 25 ms in normal speech while participants were recruited by the Department of the corresponding distributions overlap in PS. Neurology at the Hospital of Aix en Provence. All were diagnosed as having mild to moderate the Praat program. Measurements were made on idiopathic PD; they had no histories of combined wideband spectrograms and neurological, respiratory, laryngeal, speech and oscillograms displayed on a screen, and by voice diseases or disorders, apart from those listening to selected segments of the waveform in associated with PD. They were being treated with regions of specific interest. In case of single L-Dopa and were experiencing motor fluctuations intervocalic O’s, three events were identified in response to their treatment, they had adequate when present: F2-vowel offset (which coincides vision with corrective lenses and claimed not to with the beginning of the silence (voiceless O’s) suffer from hearing loss. In order to make the or the voiced bar (voiced O’s), occlusion release effects of PD more discernable, antiparkinsonian and F2-vowel onset. When O was first element of medications were withheld overnight and the first a cluster (OC2 or O.C2), the boundary between O recordings started after at least 10 hours without and C2 was at the end of the burst (if any). When medication. Before recording, the motor disability both C2 and C1 were O’s and there was no burst, of each patient was assessed using the Unified the boundary was the discontinuity (if visible) Parkinson's Disease Rating Scale (UPDRS), between the two occlusions; when C2 was a especially dysarthria severity as defined by item fricative the boundary coincided with the end of 18 (Fahn S. et al., 1987). The recordings were occlusion and the beginning of noise. Finally, made in conformity with the rules defined by the when C2 was a sonorant, the boundary was at the local Ethics committee. The nature of the study beginning of the sonorant-F2. When O was in the was carefully explained to each patient and second position and C1 a fricative or a sonorant, control speaker and they signed a written consent the beginning of O coincided with the complete form prior to being recorded. The eight control disappearance of noise or formants. speakers were non-neurologically impaired and had adequate vision with reading spectacles as 2.5. Reduction and assimilation analysis. appropriate and did not report hearing problems. 2.5.1.Intervocalic sound energy 2.2. Speech sample and recording equipment Speech sound weakening correlates with a The read speech sample was a paragraph of La reduced energy of realisations of consonants and Chèvre de Monsieur Seguin [29]. Each subject vowels [4,5]. However, since the modulation of was asked to read at his habitual speech rate. The sound energy conveys the information, not the text was written on paper and held in front of absolute level, the overall variation due to PD was subjects by a research assistant. High-quality obtained by measuring the relative sound energy recordings were obtained in a sound-treated room of O’s with respect to their flanking vowels. The of the Aix-en-Provence Hospital. The acoustic intervocalic sound energy difference was defined signal was transduced using an AKG C410 head as (V1mean -C mean +V2 mean -C mean)/2. The mean mounted microphone and recorded directly onto a values were extracted automatically and the PC hard disk at a sampling rate of 20 KHz. results were examined for each target O in relation to voicing and place and compared between PS 2.3.Transcriptions. and CS. The author transcribed readings orthographically 2.5.2. Duration. and O’s were identified in the text. The number of O’s examined in the study can be seen in Table 1. The durations of O’s, occlusions and VOT’s were Table 1. Number of analysed O’s in PS and CS (Lab:labial, extracted automatically from the labelled files. den: dental, Vel: velar) 2.5.3.Voicing. Voiceless Voiced Lab Den Vel Lab Den Vel Voicing is difficult to define because there is a p t k b d g great deal of variation in how it is manifested PS 144 219 144 102 209 38 during stop closures. Voicing may remain light CS 121 197 121 87 172 33 throughout the O but can also occur only in the first third and/or last third of the C. Voicing . patterns were reported for each O. 2..4. Labeling criteria 2.5.4. Other characteristics of weakened O’s. Acoustic measures were obtained by hand, using O-weakening may result in the loss of certain /b/’s and /d/’s are longer in PS than in CS. All features such as burst and occlusion. Therefore, voiceless O’s are shorter in PS although the presence of an occlusion and a burst was differences are only significant for /t/’s. checked for each O. Some O’s may be changed into fricatives, sonorants or approximants and 3.2.2. VOT duration exhibit noise or formants. The presence of such cues was checked for each O. In the vicinity of Table 4. VOT duration (in ms ) and standard deviation (SD) nasal vowels O’s may be partially or totally in PS and CS. T-values for each variable exceeding the 0.05 alpha-level are indicated by an asterisk. nasalised and exhibit mid-frequency formants. Two patterns of nasalisation (N) were determined: p t k b d g partial N (a separate O exhibiting an occlusion PS 25 42 38 12 20 15 (12) (19) (13) (6) (15) (6) and/or burst) and total N (no interruption of mid- CS 22 44 37 10 13 19 frequency formants).