MFCC-based Recurrent Neural Network for automatic clinical depression recognition and assessment from speech

Rejaibi, Emna; Komaty, Ali; Mériaudeau, Fabrice; Agrebi, Said; Othmani, Alice

doi:10.1016/j.bspc.2021.103107

Cited by 136 publications

(49 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The authors of this paper acknowledge that higher baseline F1-scores for the test set have been reported in [31,14]. However, a comparison cannot be made because, unlike the approach in this paper, those systems either employed a different evaluation protocol such as segment-level predictions or re-partitioned the train-validation-test splits.…”

Section: Multi Frame Rate Trainingmentioning

confidence: 98%

“…In this work, DepAudioNet was chosen as a baseline mainly because of DepAudioNet's open-source code [24]. The original paper where DepAudioNet was first proposed [9] did not report test-set results because of the unavailability of ground truth labels and computed the F1-score for predictions at the speaker level and not at the frame level [14] nor the segment level [31]. Unfortunately, there is a lack of consensus regarding the evaluation protocols for the DAIC-WOZ dataset.…”

Section: Multi Frame Rate Trainingmentioning

confidence: 99%

“…Recently, speech-based automatic diagnosis of depression has gained significant momentum [6,7,8] and advancements in deep learning have pushed their performance to newer heights [9,10,11,12,13,14]. However, data scarcity still remains one of the major challenges in building reliable systems for MDD modeling purposes.…”

Section: Introductionmentioning

confidence: 99%

“…However, GANs themselves require significant amount of training data to be effective. In [10], spectrograms were rotated to generate new samples and in [14], noise, pitch-shifting and speed perturbation was employed to augment training data. In [13], a multi-window data augmentation was proposed for emotion recognition which used multiple frame-widths.…”

Section: Introductionmentioning

confidence: 99%

“…In [13], a multi-window data augmentation was proposed for emotion recognition which used multiple frame-widths. However, the methods proposed in [10,13,14] were not compared with conventional data augmentation techniques and were only evaluated using one model.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

FrAUG: A Frame Rate Based Data Augmentation Method for Depression Detection from Speech Signals

Ravi¹,

Wang²,

Flint³

et al. 2022

Preprint

View full text Add to dashboard Cite

In this paper, a data augmentation method is proposed for depression detection from speech signals. Samples for data augmentation were created by changing the frame-width and the frame-shift parameters during the feature extraction process. Unlike other data augmentation methods (such as VTLP, pitch perturbation, or speed perturbation), the proposed method does not explicitly change acoustic parameters but rather the time-frequency resolution of frame-level features. The proposed method was evaluated using two different datasets, models, and input acoustic features. For the DAIC-WOZ (English) dataset when using the DepAudioNet model and mel-Spectrograms as input, the proposed method resulted in an improvement of 5.97% (validation) and 25.13% (test) when compared to the baseline. The improvements for the CON-VERGE (Mandarin) dataset when using the x-vector embeddings with CNN as the backend and MFCCs as input features were 9.32% (validation) and 12.99% (test). Baseline systems do not incorporate any data augmentation. Further, the proposed method outperformed commonly used data-augmentation methods such as noise augmentation, VTLP, Speed, and Pitch Perturbation. All improvements were statistically significant.

show abstract

Section: Multi Frame Rate Trainingmentioning

confidence: 98%

Section: Multi Frame Rate Trainingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

FrAUG: A Frame Rate Based Data Augmentation Method for Depression Detection from Speech Signals

Ravi¹,

Wang²,

Flint³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

A Proposal for Developing and Deploying Statistical Dialog Management in Commercial Conversational Platforms

Cañas

Griol²,

Callejas³

2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Conversational interfaces have recently become ubiquitous in the personal sphere by improving an individual's quality of life and industrial environments by automating services and their corresponding cost savings. However, designing the dialog model used by these interfaces to decide the following response is a hard-to-accomplish task for complex conversational interactions. This paper proposes a statistical-based dialog manager architecture, which provides flexibility to develop and maintain this module. Our proposal has been integrated using DialogFlow, a natural language understanding platform provided by Google to design conversational user interfaces. The proposed hybrid architecture has been assessed with a real use case for a train scheduling domain, proving that the user experience is highly valued and can be integrated into commercial setups.Conversational Interfaces (CUI) are systems that simulate interactive conversations with humans [1, 2, 3]. These systems are meant to display human-like characteristics and support the use of spontaneous natural language to interact with different purposes, such as performing transactions, responding to questions, or chatting.These interfaces have become a key research subject for many companies, as they have understood the potential revenue of introducing these devices into society's mainstream. Virtual Personal Assistants (VPAs), such as Google Now, Apple's Siri, Amazon's Alexa, or Microsoft's Cortana, are the most representative examples. These VPAs are used for a wide variety of tasks, from setting the alarm to updating the calendar, passing through getting directions, finding the nearest restaurants or stores, or even planning a recipe or reporting the news. CUIs are also used to make their services more engaging and increase customer satisfaction in various applications, such as making appointments and reservations, answering questions, e-government procedures, and support services.On the other hand, machine learning and deep learning methodologies have become part of many of the current intelligent systems [4,5,6]. These methodologies have traditionally been used in the field of conversational interfaces to develop automatic speech recognizers [7,8] and, more recently, to implement natural language understanding modules, natural language generators, and dialog management processes [4,9,10].

show abstract

Detection of Speech Related Disorders by Pre-trained Embedding Models Extracted Biomarkers

Jenei

Kiss

Schiferl

2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

MFCC-based Recurrent Neural Network for automatic clinical depression recognition and assessment from speech

Cited by 136 publications

References 24 publications

FrAUG: A Frame Rate Based Data Augmentation Method for Depression Detection from Speech Signals

FrAUG: A Frame Rate Based Data Augmentation Method for Depression Detection from Speech Signals

A Proposal for Developing and Deploying Statistical Dialog Management in Commercial Conversational Platforms

Detection of Speech Related Disorders by Pre-trained Embedding Models Extracted Biomarkers

Contact Info

Product

Resources

About