Language Independent Assessment of Motor Impairments of Patients with Parkinson’s Disease Using i-Vectors

Garcia, N.; Vásquez-Correa, Juan Camilo; Orozco-Arroyave, Juan Rafael; Dehak, Najim; Nöth, Elmar

doi:10.1007/978-3-319-64206-2_17

Cited by 3 publications

(3 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…When not enough speech data is available to train the GMM models, which mainly occurs when GMMs are used to model each subject (rather than a group), GMMs can be adapted from Universal Background Models (UBM) previously trained with a bigger dataset (Reynolds et al, 2000;Bocklet et al, 2013). More than that, a more recent speaker recognition technique, called i-vectors, has been adapted for PD detection (Garcia et al, 2017;Moro-Velázquez et al, 2018). This approach consists in removing the UBM mean supervector and projecting each supervector onto a lower dimensional space, called the total variability space.…”

Section: Introductionmentioning

confidence: 99%

X-Vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection From Speech

Jeancolas

Petrovska‐Delacrétaz

Mangone

et al. 2021

Front. Neuroinform.

View full text Add to dashboard Cite

Many articles have used voice analysis to detect Parkinson's disease (PD), but few have focused on the early stages of the disease and the gender effect. In this article, we have adapted the latest speaker recognition system, called x-vectors, in order to detect PD at an early stage using voice analysis. X-vectors are embeddings extracted from Deep Neural Networks (DNNs), which provide robust speaker representations and improve speaker recognition when large amounts of training data are used. Our goal was to assess whether, in the context of early PD detection, this technique would outperform the more standard classifier MFCC-GMM (Mel-Frequency Cepstral Coefficients—Gaussian Mixture Model) and, if so, under which conditions. We recorded 221 French speakers (recently diagnosed PD subjects and healthy controls) with a high-quality microphone and via the telephone network. Men and women were analyzed separately in order to have more precise models and to assess a possible gender effect. Several experimental and methodological aspects were tested in order to analyze their impacts on classification performance. We assessed the impact of the audio segment durations, data augmentation, type of dataset used for the neural network training, kind of speech tasks, and back-end analyses. X-vectors technique provided better classification performances than MFCC-GMM for the text-independent tasks, and seemed to be particularly suited for the early detection of PD in women (7–15% improvement). This result was observed for both recording types (high-quality microphone and telephone).

show abstract

Section: Introductionmentioning

confidence: 99%

X-Vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection From Speech

Jeancolas

Petrovska‐Delacrétaz

Mangone

et al. 2021

Front. Neuroinform.

View full text Add to dashboard Cite

show abstract

“…Recently, in [24] the i-vector approach was applied to assess the neurological state of a group with 50 PD patients. Similarly, in [25] speech impairments of PD patients speaking three different languages (Spanish, German, and Czech) were evaluated considering the i-vector approach. The results indicate that this method is suitable to be applied in different languages.…”

Section: Longitudinal Monitoring Of Pd From Speechmentioning

confidence: 99%

Speaker models for monitoring Parkinson’s disease progression considering different communication channels and acoustic conditions

Arias-Vergara

Vásquez-Correa

Orozco-Arroyave

et al. 2018

Speech Communication

Self Cite

View full text Add to dashboard Cite

The interest of the research community in the analysis of speech of people suffering from Parkinson's disease has increased in recent years. Most of the studies are focused on developing computer-aided tools for the detection and unobtrusive monitoring the progression of several symptoms of the disease. Different approaches have been proposed to detect several voice impairments in PD patients. Most of the state-of-the-art studies address the task of assessing the neurological state of patients considering information obtained from a set of PD speakers, i.e., most of the reported studies consider regression techniques which are trained with information obtained from a group of patients and assess the neurological state of another group of patients suffering from PD. Such an approach seems to be a good alternative to evaluate the suitability of the models/measures extracted from the speech signals; however, that approach is not appropriate to perform individual monitoring of patients and including information about the progression of the disease in a specific person. Additionally, due to the difficulty of having continue access to PD patients, the number of contributions focused on the automatic monitoring of the patients is reduced. Most of the reported works are based on recordings captured during clinical appointments, i.e., relatively controlled acoustic and recording conditions. In this study we propose a methodology to assess the disease progression considering individual information per patient, i.e., individual speaker models. Two different methods are explored, one is based on the GMM-UBM approach and the other one is based on i-vectors. Both approaches have been successfully applied in speaker identification and verification tasks. In this paper the main hypothesis is that once the speech of a patient is accurately modeled, any change, like those that appear due to the disease progression, will be detected. Speech signals are modeled considering three speech aspects: phonation, articulation, and prosody. The results obtained with the proposed approaches are compared with respect to the traditional framework which is based on regression analysis. The models are trained considering a set with 100 speakers (50 suffering from PD and 50 healthy speakers). The tests are performed considering two sets with speech recordings 1 captured in real-world acoustic conditions. The first set contains a group of seven speakers recorded several times from 2012 to 2016, i.e., longitudinal recordings. As the acoustic conditions of those recordings were different between sessions, this corpus represents a realworld scenario to study the neurological state of PD patients. The second set is form with recordings of the same group of seven patients recorded in their houses, i.e., at-home recordings, those patients were recorded in 16 sessions during four months, i.e., one day per month, every two hours during eight hours per day. As in the case of the longitudinal recordings, the acoustic conditions were not controlled, thus this...

show abstract

“…When not enough speech data is available to train the GMM models, which mainly occurs when GMM are used to model each subject (rather than a group), GMM can be adapted from Universal Background Models (UBM) previously trained with a bigger dataset [49,5]. More than that, a more recent speaker recognition technique, called ivectors, has been adapted for PD detection [17,38]. This approach consists in removing the UBM mean supervector and projecting each supervector onto a low dimensional space, called total variability space.…”

Section: Introductionmentioning

confidence: 99%

X-vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection from Speech

Jeancolas,

Petrovska-Delacrétaz,

Mangone

et al. 2020

Preprint

View full text Add to dashboard Cite

Many articles have used voice analysis to detectParkinson's disease (PD), but few have focused on the early stages of the disease and the gender effect. In this article, we have adapted the latest speaker recognition system, called x-vectors, in order to detect an early stage of PD from voice analysis. X-vectors are embeddings extracted from a deep neural network, which provide robust speaker representations and improve speaker recognition when large amounts of training data are used. Our goal was to assess whether, in the context of early PD detection, this technique would outperform the more standard classifier MFCC-GMM (Mel-Frequency Cepstral Coefficients -Gaussian Mixture Model) and, if so, under which conditions. We recorded 221 French speakers (including recently diagnosed PD subjects and healthy controls) with a high-quality microphone and with their own telephone. Men and women were analyzed separately in order to have more precise models and to assess a possible gender effect. Several experimental and methodological aspects were tested in order to analyze their impacts on classification performance. We assessed the impact of audio segment duration, data augmentation, type of dataset used for the neural network training, kind of speech tasks, and back-end analyses. X-vectors technique provided better classification performances than MFCC-GMM for text-independent tasks, and seemed to be particularly suited for the early detection of PD in women (7 to 15% improvement). This result was observed for both recording types (high-quality microphone and telephone).

show abstract

Language Independent Assessment of Motor Impairments of Patients with Parkinson’s Disease Using i-Vectors

Cited by 3 publications

References 17 publications

X-Vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection From Speech

X-Vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection From Speech

Speaker models for monitoring Parkinson’s disease progression considering different communication channels and acoustic conditions

X-vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection from Speech

Contact Info

Product

Resources

About