scispace - formally typeset
Search or ask a question
Author

Youxiang Zhu

Bio: Youxiang Zhu is an academic researcher from University of Massachusetts Boston. The author has contributed to research in topics: Dementia & Computer science. The author has an hindex of 2, co-authored 5 publications receiving 6 citations.

Papers
More filters
Proceedings ArticleDOI
30 Aug 2021
TL;DR: This paper determines the locations and lengths of inter-word pauses using the number of blank tokens from Wav2vec where the threshold for setting the pauses is automatically generated via BERT, enabling the fine-tuning of WavBERT with non-semantic information.
Abstract: In this paper, we exploit semantic and non-semantic information from patient’s speech data using Wav2vec and Bidirectional Encoder Representations from Transformers (BERT) for dementia detection. We first propose a basic WavBERT model by extracting semantic information from speech data using Wav2vec, and analyzing the semantic information using BERT for dementia detection. While the basic model discards the non-semantic information, we propose extended WavBERT models that convert the output of Wav2vec to the input to BERT for preserving the non-semantic information in dementia detection. Specifically, we determine the locations and lengths of inter-word pauses using the number of blank tokens from Wav2vec where the threshold for setting the pauses is automatically generated via BERT. We further design a pre-trained embedding conversion network that converts the output embedding of Wav2vec to the input embedding of BERT, enabling the fine-tuning of WavBERT with non-semantic information. Our evaluation results using the ADReSSo dataset showed that the WavBERT models achieved the highest accuracy of 83.1% in the classification task, the lowest Root-Mean-Square Error (RMSE) score of 4.44 in the regression task, and a mean F1 of 70.91% in the progression task. We confirmed the effectiveness of WavBERT models exploiting both semantic and non-semantic speech.

18 citations

Journal ArticleDOI
TL;DR: In this article, a variety of transfer learning models using commonly employed MobileNet (image), YAMNet (audio), Mockingjay (speech), and BERT (text) models were compared.
Abstract: Examination of speech datasets for detecting dementia, collected via various speech tasks, has revealed links between speech and cognitive abilities. However, the speech dataset available for this research is extremely limited because the collection process of speech and baseline data from patients with dementia in clinical settings is expensive. In this paper, we study the spontaneous speech dataset from a recent ADReSS challenge, a Cookie Theft Picture (CTP) dataset with balanced groups of participants in age, gender, and cognitive status. We explore state-of-the-art deep transfer learning techniques from image, audio, speech, and language domains. We envision that one advantage of transfer learning is to eliminate the design of handcrafted features based on the tasks and datasets. Transfer learning further mitigates the limited dementia-relevant speech data problem by inheriting knowledge from similar but much larger datasets. Specifically, we built a variety of transfer learning models using commonly employed MobileNet (image), YAMNet (audio), Mockingjay (speech), and BERT (text) models. Results indicated that the transfer learning models of text data showed significantly better performance than those of audio data. Performance gains of the text models may be due to the high similarity between the pre-training text dataset and the CTP text dataset. Our multi-modal transfer learning introduced a slight improvement in accuracy, demonstrating that audio and text data provide limited complementary information. Multi-task transfer learning resulted in limited improvements in classification and a negative impact in regression. By analyzing the meaning behind the AD/non-AD labels and Mini-Mental State Examination (MMSE) scores, we observed that the inconsistency between labels and scores could limit the performance of the multi-task learning, especially when the outputs of the single-task models are highly consistent with the corresponding labels/scores. In sum, we conducted a large comparative analysis of varying transfer learning models focusing less on model customization but more on pre-trained models and pre-training datasets. We revealed insightful relations among models, data types, and data labels in this research area.

14 citations

Journal ArticleDOI
TL;DR: This paper explores the voice commands using a Voice-Assistant System (VAS), i.e., Amazon Alexa, from 40 older adults who were either Healthy Control (HC) participants or Mild Cognitive Impairment (MCI) participants, age 65 or older, to demonstrate the promise of future home-based cognitive assessments using Voice- Assistant Systems.

9 citations

Posted Content
TL;DR: This paper exploits a Fully Convolutional Network to analyze the audio data of spontaneous speech for dementia detection, and builds a convolutional layer to produce a heatmap using Otsu's method for visualization, enabling to understand the impact of the time-series audio segments on the classification results.
Abstract: In this paper, we exploit a Fully Convolutional Network (FCN) to analyze the audio data of spontaneous speech for dementia detection. A fully convolutional network accommodates speech samples with varying lengths, thus enabling us to analyze the speech sample without manual segmentation. Specifically, we first obtain the Mel Frequency Cepstral Coefficient (MFCC) feature map from each participant's audio data and convert the speech classification task on audio data to an image classification task on MFCC feature maps. Then, to solve the data insufficiency problem, we apply transfer learning by adopting a pre-trained backbone Convolutional Neural Network (CNN) model from the MobileNet architecture and the ImageNet dataset. We further build a convolutional layer to produce a heatmap using Otsu's method for visualization, enabling us to understand the impact of the time-series audio segments on the classification results. We demonstrate that our classification model achieves 66.7% over the testing dataset, 62.5% of the baseline model provided in the ADReSS challenge. Through the visualization technique, we can evaluate the impact of audio segments, such as filled pauses from the participants and repeated questions from the investigator, on the classification results.

2 citations

Posted Content
TL;DR: In this paper, the authors used adversarial training to evaluate the impact of the perturbation in training samples on the detection model and found that some pauses are more sensitive to dementia than other pauses from the model's perspective, e.g. near to the verb "is".
Abstract: Speech pause is an effective biomarker in dementia detection. Recent deep learning models have exploited speech pauses to achieve highly accurate dementia detection, but have not exploited the interpretability of speech pauses, i.e., what and how positions and lengths of speech pauses affect the result of dementia detection. In this paper, we will study the positions and lengths of dementia-sensitive pauses using adversarial learning approaches. Specifically, we first utilize an adversarial attack approach by adding the perturbation to the speech pauses of the testing samples, aiming to reduce the confidence levels of the detection model. Then, we apply an adversarial training approach to evaluate the impact of the perturbation in training samples on the detection model. We examine the interpretability from the perspectives of model accuracy, pause context, and pause length. We found that some pauses are more sensitive to dementia than other pauses from the model's perspective, e.g., speech pauses near to the verb "is". Increasing lengths of sensitive pauses or adding sensitive pauses leads the model inference to Alzheimer's Disease, while decreasing the lengths of sensitive pauses or deleting sensitive pauses leads to non-AD.

1 citations


Cited by
More filters
Proceedings ArticleDOI
30 Aug 2021
TL;DR: This paper determines the locations and lengths of inter-word pauses using the number of blank tokens from Wav2vec where the threshold for setting the pauses is automatically generated via BERT, enabling the fine-tuning of WavBERT with non-semantic information.
Abstract: In this paper, we exploit semantic and non-semantic information from patient’s speech data using Wav2vec and Bidirectional Encoder Representations from Transformers (BERT) for dementia detection. We first propose a basic WavBERT model by extracting semantic information from speech data using Wav2vec, and analyzing the semantic information using BERT for dementia detection. While the basic model discards the non-semantic information, we propose extended WavBERT models that convert the output of Wav2vec to the input to BERT for preserving the non-semantic information in dementia detection. Specifically, we determine the locations and lengths of inter-word pauses using the number of blank tokens from Wav2vec where the threshold for setting the pauses is automatically generated via BERT. We further design a pre-trained embedding conversion network that converts the output embedding of Wav2vec to the input embedding of BERT, enabling the fine-tuning of WavBERT with non-semantic information. Our evaluation results using the ADReSSo dataset showed that the WavBERT models achieved the highest accuracy of 83.1% in the classification task, the lowest Root-Mean-Square Error (RMSE) score of 4.44 in the regression task, and a mean F1 of 70.91% in the progression task. We confirmed the effectiveness of WavBERT models exploiting both semantic and non-semantic speech.

18 citations

Journal ArticleDOI
TL;DR: In this article , a combination of BERT, Vision Transformer, Co-Attention, Multimodal Shifting Gate, and a variant of the self-attention mechanism was proposed to detect Alzheimer's dementia and predict the Mini-Mental State Examination (MMSE) scores.
Abstract: Alzheimer's dementia (AD) entails negative psychological, social, and economic consequences not only for the patients but also for their families, relatives, and society in general. Despite the significance of this phenomenon and the importance for an early diagnosis, there are still limitations. Specifically, the main limitation is pertinent to the way the modalities of speech and transcripts are combined in a single neural network. Existing research works add/concatenate the image and text representations, employ majority voting approaches or average the predictions after training many textual and speech models separately. To address these limitations, in this article we present some new methods to detect AD patients and predict the Mini-Mental State Examination (MMSE) scores in an end-to-end trainable manner consisting of a combination of BERT, Vision Transformer, Co-Attention, Multimodal Shifting Gate, and a variant of the self-attention mechanism. Specifically, we convert audio to Log-Mel spectrograms, their delta, and delta-delta (acceleration values). First, we pass each transcript and image through a BERT model and Vision Transformer, respectively, adding a co-attention layer at the top, which generates image and word attention simultaneously. Secondly, we propose an architecture, which integrates multimodal information to a BERT model via a Multimodal Shifting Gate. Finally, we introduce an approach to capture both the inter- and intra-modal interactions by concatenating the textual and visual representations and utilizing a self-attention mechanism, which includes a gate model. Experiments conducted on the ADReSS Challenge dataset indicate that our introduced models demonstrate valuable advantages over existing research initiatives achieving competitive results in both the AD classification and MMSE regression task. Specifically, our best performing model attains an accuracy of 90.00% and a Root Mean Squared Error (RMSE) of 3.61 in the AD classification task and MMSE regression task, respectively, achieving a new state-of-the-art performance in the MMSE regression task.

12 citations

Proceedings ArticleDOI
04 Apr 2022
TL;DR: In this article , the authors explored the usefulness of using Wav2Vec self-supervised speech representations as features for training an ASR system for dysarthric speech.
Abstract: State-of-the-art automatic speech recognition (ASR) systems perform well on healthy speech. However, the performance on impaired speech still remains an issue. The current study explores the usefulness of using Wav2Vec self-supervised speech representations as features for training an ASR system for dysarthric speech. Dysarthric speech recognition is particu-larly difficult as several aspects of speech such as articulation, prosody and phonation can be impaired. Specifically, we train an acoustic model with features extracted from Wav2Vec, Hubert, and the cross-lingual XLSR model. Results suggest that speech representations pretrained on large unlabelled data can improve word error rate (WER) performance. In particular, features from the multilingual model led to lower WERs than filter-banks (Fbank) or models trained on a single language. Improvements were observed in English speakers with cerebral palsy caused dysarthria (UASpeech corpus), Spanish speakers with Parkinsonian dysarthria (PC-GITA corpus) and Italian speakers with paralysis-based dysarthria (EasyCall corpus). Compared to using Fbank features, XLSR-based features reduced WERs by 6.8%, 22.0%, and 7.0% for the UASpeech, PC-GITA, and EasyCall corpus, respectively.

11 citations

Proceedings ArticleDOI
23 Jun 2022
TL;DR: A state-of-the-art Conformer based speech recognition system built on the DementiaBank Pitt corpus for automatic AD detection is presented, with the best-published speech recognition based AD detection accuracy of 91.7% obtained.
Abstract: Early diagnosis of Alzheimer’s disease (AD) is crucial in facilitating preventive care to delay further progression. This paper presents the development of a state-of-the-art Conformer based speech recognition system built on the DementiaBank Pitt corpus for automatic AD detection. The baseline Conformer system trained with speed perturbation and SpecAugment based data augmentation is significantly improved by incorporating a set of purposefully designed modeling features, including neural architecture search based auto-configuration of domain-specific Conformer hyper-parameters in addition to parameter fine-tuning; fine-grained elderly speaker adaptation using learning hidden unit contributions (LHUC); and two-pass cross-system rescoring based combination with hybrid TDNN systems. An overall word error rate (WER) reduction of 13.6% absolute (34.8% relative) was obtained on the evaluation data of 48 elderly speakers. Using the final systems’ recognition outputs to extract textual features, the best-published speech recognition based AD detection accuracy of 91.7% was obtained.

8 citations

Journal ArticleDOI
TL;DR: The transfer learning method in this study improves AD prediction, which does not only reduces the need for feature engineering but also addresses the lack of sufficiently large datasets.
Abstract: Alzheimer's disease (AD) is a neurodegenerative disease that is difficult to be detected using convenient and reliable methods. The language change in patients with AD is an important signal of their cognitive status, which potentially helps in early diagnosis. In this study, we developed a transfer learning model based on speech and natural language processing (NLP) technology for the early diagnosis of AD. The lack of large datasets limits the use of complex neural network models without feature engineering, while transfer learning can effectively solve this problem. The transfer learning model is firstly pre-trained on large text datasets to get the pre-trained language model, and then, based on such a model, an AD classification model is performed on small training sets. Concretely, a distilled bidirectional encoder representation (distilBert) embedding, combined with a logistic regression classifier, is used to distinguish AD from normal controls. The model experiment was evaluated on Alzheimer's dementia recognition through spontaneous speech datasets in 2020, including the balanced 78 healthy controls (HC) and 78 patients with AD. The accuracy of the proposed model is 0.88, which is almost equivalent to the champion score in the challenge and a considerable improvement over the baseline of 75% established by organizers of the challenge. As a result, the transfer learning method in this study improves AD prediction, which does not only reduces the need for feature engineering but also addresses the lack of sufficiently large datasets.

8 citations