Characterisation of voice quality of Parkinson’s disease using differential phonological posterior features

TLDR: The proposed characterisation of VQ might also be applied to other kinds of pathological speech, and pathological voice quality is characterised using healthy non-modal voice quality “base/eigenspace”.

About: This article is published in Computer Speech & Language.The article was published on 2017-11-01 and is currently open access. It has received 50 citations till now. The article focuses on the topics: Voice analysis & Creaky voice.

Figures

Table 3: Euclidean distances ql between non-modal and Parkinsonian DPP features. As Euclidean distance is a similarity measure, whereby smaller is more similar, we calculate an inverse of the Euclidean distance to plot composition of non-modal voice quality in Parkinsonian speech in Figure 2.

Table 2: The impact of non-modal phonation on phonological features, measured as a positive (+) or negative (−) difference between the mean phonological posteriors of speech with modal phonation, and the mean phonological posteriors with non-modal phonation. The three features with the greatest differences are listed. Invariance is concluded based on statistics in Table 1.

Table 4: The Spearman’s correlation coefficients between the real and predicted modified FDA scores related to the larynx. Median values are calculated for the correlations with the three evaluators. Results obtained for the three sub-sets of the PD data (see Section 4.2) are reported.

Figure 1: Mean modal/HC SPE posteriors µk (top-left figures) and differentials ∆µk of non-modal/PD phonations with respect to the modal/HC voice.

Figure 2: Composition of voice quality in Parkinson’s speech.

Frequently asked questions

Q1. What are the contributions in "Characterisation of voice quality of parkinson’s disease using differential phonological posterior features" ?

A goal of this paper is to characterize a VQ spectrum – the composition of non-modal phonations – of voice in PD. The paper relates non-modal healthy phonations: breathy, creaky, tense, falsetto and harsh, with disordered phonation in PD. In addition, the proposed features were applied for prediction ∗Corresponding author Email address: milos. 

Q2. What future works have the authors mentioned in the paper "Characterisation of voice quality of parkinson’s disease using differential phonological posterior features" ?

In future, the authors plan to obtain PD data with labeled VQ, and validate the VQ characterisation on individual patients, looking for example for regression of the perceptual scores. 

Q3. What type of phonation type was used for the read-VQ database?

The read-VQ database contains 4 speakers (2 males and 2 females) who were asked to read 17 sentences in six different phonation types: modal, breathy, tense, harsh, creaky, and falsetto. 

Q4. What equipment was used to record the audio of the read-VQ database?

Audio of the read-VQ database was recorded at 44.1 kHz using high quality recording equipment: a B&K 4191 free-field microphone and a B&K 7749 pre-amplifier. 

Q5. What is the common clinical assessment method for disordered vocal quality?

Auditory-perceptual evaluation of disordered VQ is the most commonly used clinical assessment method, and is considered by clinicians as the “gold standard” for documenting voice impairment severity (Kreiman et al., 1993). 

Q6. What is the definition of hypokinetic dysarthria?

Speech of hypokinetic dysarthria in Parkinson’s disease (PD) is characterised by hypokinesia (rigid, less motion describing decreased range and frequency of movement) of the vocal folds and articulators. 

Q7. How many frames were used for the training?

The temporal context from 7 to 11 successive frames was tested with no particular performance increase, so the temporal context of 9 frames was used for the training. 

Q8. Who produced the prototype voice quality examples?

Participants were given prototype voice quality examples, produced by John Laver and John Kane, and were asked to practise producing them before coming to the recording session. 

Q9. What is the biggest open-source database for PD?

To the authors’ knowledge, the used PD database is the biggest open-source database available, containing both isolated and connected speech, and was selected primarily for its size. 

Q10. What is the phonological analysis and synthesis?

the platform is based on cascaded speech analysis and synthesis that works internally with the phonological speech representation. 

Q11. How many sentences were chosen to obtain a wide phonetic coverage?

451 sentences were chosen to obtain a wide phonetic coverage, as it is likely that it can be very difficult for speakers to maintain a constant type of phonation over a long utterance. 

Q12. What are the improvements for the monologue and reading speech tasks?

Improvements are obtained for the monologue and reading speech tasks, of 3% and 16% respectively, whereas no improvement is obtained with the pataka speech task. 

Q13. How are the results of the proposed characterisation of speech quality?

The results obtained by DPP features have been validated by matching the obtained most significant, non-modal phonation types with evaluating parameters of the perceptual assessments.