[Paper Review] Voiceprint recognition of Parkinson patients based on deep learning

Zhijing Xu, Juan Wang+2 more|arXiv (Cornell University)|Jan 1, 2018

Voice and Speech Disorders1 citations

TL;DR

This study proposes a deep learning approach combining Weighted Mel Frequency Cepstrum Coefficients (WMFCC) for voiceprint feature extraction and a Deep Neural Network (DNN) with Mini-Batch Gradient Descent (MBGD) for classifying Parkinson's disease (PD) patients from healthy individuals. The method achieves 89.5% accuracy in distinguishing PD patients using sustained vowel recordings, outperforming conventional methods like SVM.

ABSTRACT

More than 90% of the Parkinson Disease (PD) patients suffer from vocal disorders. Speech impairment is already indicator of PD. This study focuses on PD diagnosis through voiceprint features. In this paper, a method based on Deep Neural Network (DNN) recognition and classification combined with Mini-Batch Gradient Descent (MBGD) is proposed to distinguish PD patients from healthy people using voiceprint features. In order to exact the voiceprint features from patients, Weighted Mel Frequency Cepstrum Coefficients (WMFCC) is applied. The proposed method is tested on experimental data obtained by the voice recordings of three sustained vowels /a/, /o/ and /u/ from participants (48 PD and 20 healthy people). The results show that the proposed method achieves a high accuracy of diagnosis of PD patients from healthy people, than the conventional methods like Support Vector Machine (SVM) and other mentioned in this paper. The accuracy achieved is 89.5%. WMFCC approach can solve the problem that the high-order cepstrum coefficients are small and the features component's representation ability to the audio is weak. MBGD reduces the computational loads of the loss function, and increases the training speed of the system. DNN classifier enhances the classification ability of voiceprint features. Therefore, the above approaches can provide a solid solution for the quick auxiliary diagnosis of PD in early stage.

Motivation & Objective

• To address the challenge of early-stage Parkinson's disease (PD) diagnosis using non-invasive voice-based biomarkers.
• To improve the classification accuracy of PD versus healthy individuals using voiceprint features.
• To overcome limitations in traditional feature extraction methods, such as weak representation of high-order cepstrum coefficients.
• To reduce computational load and accelerate training through Mini-Batch Gradient Descent in a deep learning framework.

Proposed method

• Weighted Mel Frequency Cepstrum Coefficients (WMFCC) are used to extract voiceprint features from sustained vowels /a/, /o/, and /u/.
• The WMFCC method enhances feature representation by addressing the issue of small high-order cepstrum coefficients.
• A Deep Neural Network (DNN) classifier is trained using Mini-Batch Gradient Descent (MBGD) to improve classification performance.
• MBGD reduces computational load and accelerates convergence during DNN training.
• The system is trained and tested on a dataset of 48 PD patients and 20 healthy individuals using sustained vowel recordings.
• The DNN model learns hierarchical representations of voiceprint features to distinguish PD from healthy controls.

Experimental results

Research questions

• RQ1 Can WMFCC improve the discriminative power of voiceprint features for Parkinson’s disease detection compared to conventional MFCC?
• RQ2 How does a DNN with MBGD compare to traditional classifiers like SVM in diagnosing PD from voice samples?
• RQ3 To what extent does the proposed method reduce computational load while maintaining high diagnostic accuracy?
• RQ4 Can sustained vowel phonations reliably support early-stage PD diagnosis using deep learning?

Key findings

• The proposed method achieves a diagnostic accuracy of 89.5% in distinguishing Parkinson’s disease patients from healthy individuals.
• The use of WMFCC significantly improves feature representation, especially for high-order cepstrum coefficients.
• Mini-Batch Gradient Descent reduces the computational load and increases the training speed of the DNN model.
• The DNN classifier demonstrates superior classification performance compared to conventional methods like SVM.
• The system effectively leverages sustained vowel recordings for reliable early-stage PD detection.
• The integration of WMFCC and DNN with MBGD provides a robust and efficient solution for automated PD screening.