key: cord-0464252-ptsdeg8l
authors: Deshpande, Gauri; Schuller, Bjorn W.
title: Audio, Speech, Language,&Signal Processing for COVID-19: A Comprehensive Overview
date: 2020-11-29
journal: nan
DOI: nan
sha: 719174ffc1fb466bf4fc9f816990043c9ba16808
doc_id: 464252
cord_uid: ptsdeg8l

The Coronavirus (COVID-19) pandemic has been the research focus world-wide in the year 2020. Several efforts, from collection of COVID-19 patients' data to screening them for the virus's detection are taken with rigour. A major portion of COVID-19 symptoms are related to the functioning of the respiratory system, which in-turn critically influences the human speech production system. This drives the research focus towards identifying the markers of COVID-19 in speech and other human generated audio signals. In this paper, we give an overview of the speech and other audio signal, language and general signal processing-based work done using Artificial Intelligence techniques to screen, diagnose, monitor, and spread the awareness aboutCOVID-19. We also briefly describe the research related to detect accord-ing COVID-19 symptoms carried out so far. We aspire that this collective information will be useful in developing automated systems, which can help in the context of COVID-19 using non-obtrusive and easy to use modalities such as audio, speech, and language.

It is vital to have an easy to use tool for screening, diagnosing, and monitoring the virus and its proliferation. An automated approach to detect and monitor the presence of COVID-19 or its symptoms could be developed using Artificial Intelligence (AI) based techniques. Although, AI techniques are still in the process of reaching a matured stage, they can be still be used for early detection of the symptoms, especially in the form of a self-care tool 1 www.who.int in reducing the spread, taking early care, and hence avoiding the severe conditions to propagate. As reviewed by the authors of [1] , AI based approaches using speech and other audio modalities have several opportunities in this space. Also, the authors of [2] have been among the very first to identify and collate the useful AI based techniques and the efforts taken for COVID-19, right when the pandemic spread was at its peak.

As per the WHO department of mental health and substance abuse, the current scenario of COVID-19 is susceptible to elevate the rates of stress or anxiety among individuals. Especially, lock-down and social distancing, quarantine and its after effects in the society might have adverse effects on the levels of loneliness, depression, self-harm, or suicidal behaviour. A special attention is needed for the health care providers, having to face the trauma directly and spending long working hours in such scenarios. Both physical and mental health needs have increased and require AI to provide faster and 3 easy to access solutions. Not only does identification and monitoring need digital assistance, but also the post trauma phase would need it.

As depicted in Figure 1 , we are discussing about capturing, pre-processing and applying speech and other human audio data for screening, diagnosing, monitoring and spreading the awareness of COVID-19 in this paper. Tables 1, 2, 3, 4, and 5 explain past speech and audio related work done to provide solutions for COVID-19 related health problems such as cough sound, asthmatic sound, obstructive sleep apnea, breathing rate, and stress detection from speech signals. The references given in these tables can be used readily by those who wants to develop and provide immediate solutions in this space. These tables also mention the machine learning and deep learning technologies used on the mentioned data-sets to provide mentioned accuracy.

For each speech or audio application, there is a vast space in the literature.

We have selected only a handful of them, considering their relevance in the COVID-19 situation. Not everything that can be developed, can be used in this scenario considering other factors such as social distancing and personal and environmental hygiene. Hence, the developments are required to be driven by guidance from the clinicians and health care providers.

The rest of the paper is organised as follows. We start with Section 2 to explain the recent advances towards providing COVID-19 screening and diagnosing tools using speech and audio signals. Similarly, in Section 3 we talk about the recent development towards the applications of speech and audio processing for monitoring the spread of the COVID-19 virus. In Section 4, we describe the use of speech based technologies in the growing awareness about the pandemic so that the accurate knowledge about the disease, factors 4 that control its expansion and preliminary care that one has to take is known to most of the individuals. We talk about the gaps where further research and development is required in Section 5, and we conclude our discussion in Section 6.

In the clinical test for diagnosing COVID-19 infection, the anterior nasal swabs sample is collected as suggested in [9] . These tests are performed by the healthcare providers belonging to local or state healthcare departments.

As the medicinal drug or vaccine for the treatment of COVID-19 is still not available, it is imperative to detect the infected individuals and physically separate them from the healthy community to stop its wide spread. Hence, additionally, as a part of any personal informatics system, the availability of self-screening or self-diagnosing methodologies can aid in detecting and self-isolating at an early stage itself.

There exists a fine line between screening and diagnosing, where screening gives an early indication of the presence of a disease and diagnosing confirms the presence/absence of disease. Screening is probabilistic, whereas diagnosis is binary in nature. To scale-up the detection of COVID-19 virus, we discuss different algorithms/applications using audio processing for screening and diagnosis of COVID-19 in this section.

Cough detection is about identifying the cough sound and differentiate it from other similar sounds such as speech and laughter and also, to identify COVID-19 specific cough. As a first step, it requires cough and speech samples of the same subject followed by collecting the COVID-19 and non-COVID-19 cough sound samples so as to develop an AI model that can differentiate between them on its own. Figure 2 shows the number of healthy and COVID-19 positive subjects or samples data collected by different groups.

Cough audio samples can be collected using a simple smartphone microphone. The Cambridge University 2 provided a web based platform and an android application for the general population to upload their cough sounds along with some additional information such as their age, gender, brief medical history, location, symptoms, and if the participating subjects had been tested positive. This data collection platform ask the participants to also read a sentence so that their speech can also be recorded. The samples comprise of 3 coughs, 5 breaths, and voice while reading via a periodic survey which captured this data after every 2 days. However, in the work presented in [10] , only cough and breathing sounds are considered. As explained in [10] , the crowdsourced data collected comes from more than 10 different countries and comprises of samples from more than 7 000 subjects with more than As explained in [19] , another corpus called "Coswara" of 941 subjects and 9 different sounds is formed using a web interface developed by IISC Principal Component Analysis (PCA) on audio spectrograms, Fast Fourier Transfor (FFT) coefficients, and RFC Acted cough from 17 patients having a cough due to common cold (n=8), asthma (n=3), allergies (n=1), and chronic cough (n=5) Result True positive rate: 92 %, false positive rate: 0.5 %. [15] Three spectral band features along with a logistic regression classifier 43 real-world environment recordings from the video sources of [16] , with sounds of COPD, pertussis, croup, common cold, bronchitis, bronchiolitis & asthma. Result Sensitivity: 90.31 %, specificity: 98.14 %, F1-score: 88.70 %. [17] Local image (Hu) moments over audio spectrograms 13 subjects samples with 3 respiratory conditions for 1 day having low, medium, and high noisy background. Result Smartphone based cough detector having 88.94 % sensitivity and 98.64 % specificity in noisy environments with optimal battery consumption. 10 Bangalore India. The nine sounds include, shallow and deep cough, shallow and deep breathing, the vowels /ey/, /i/, and /u/, and one to twenty digit counting in normal and fast speaking rate. The metadata collected from the participants includes age, gender, location, current health status (healthy / exposed / cured / infected) and the presence of co-morbidity. Here, along with the subjects labelling their own data, one more annotation is provided negative COVID-19 clinically tested individuals is collected in [24] . The subjects were asked to provide the speech of /ah/ and /z/, counting from 50 to 80 and coughs for 14 days. The authors have compared the performance using three deep learning components, attention based transformer, a GRUbased expert classifier with aggressive regularisation and ensemble stacking.

The authors report that the performance with /z/ phoneme is better when compared to /ah/ and counting. Among the deep learning techniques, transformer based experiments gave better F1 scores.

A cloud based smartphone app is described in [31] , for detecting COVID-19 cough. As a first step, the authors have used a CNN based cough detector, which identifies cough sounds from over 50 environmental sounds. The authors built this detector using the ESC-50 dataset [32] . In the next stage, to diagnose a COVID-19 cough, they collected 96 bronchitis, 130 pertussis, 70 COVID-19, and 247 normal cough samples to train their COVID-19 cough detector model. Using MFCCs for feature representation and t-distributed stochastic neighbour embedding for dimensionality reduction, they have trained three models: a deep transfer learning-based multi-class classifier (using a CNN), a classical machine learning-based multi-class clas- Result Markers of asthmatic individuals: Lower pitch, higher standard deviation of pitch, higher degree of voice breaks, lower intensity, shimmer value greater than 3.8, higher jitter, average Harmonics to Noise Ratio (HNR) of 14.4, higher F1, and lower F2. [28] MFCCs with Gaussian Mixture Model (GMM)

Respiratory sound of 9 asthmatic and 9 non-asthmatic adults Result Sensitivity of 0.881 and a specificity of 0.995. [29] Correlation between HNR of speech signal and Forced Expiratory Volume to Forced Vital Capacity (FEV1/FVC) ratio obtained from spirometry vocal cords, sentiments, and lung & respiratory tract, the authors of [33] have used these bio-markers as a pre-training step for the baseline model. The authors of [44] discuss the interpretability of their framework of COVID-19 diagnosis using embeddings for the cough features and symptoms' metadata. In this study, cough, breathing, and speech of counting from 1-10 is collected from 150 subjects, in which 100 subjects were COVID-19 positive, and 50 were tested negative during their RT-PCR test. Apart from this, the authors also collected data for bronchitis and asthma cough from online and offline sources. The authors report an improvement of 5-6 % in the accuracy, F1-score, sensitivity, specificity, and precision when using both the symptoms' metadata and cough features for the classification tasks. Similarly, the authors of [45] have used additional information such as: airflow from spirometer, body temperature from thermal camera, heart rate from ECG, electrical activity that causes muscle contraction around the heart and chest region, along with cough detection from a microphone to build a sensor-based system for identifying COVID-19 subjects.

As discussed before, shortness of breath is also one of the symptoms of the virus for which smartphone apps are designed to capture breathing patterns by recording the speech signal. As described in Section 2.1 and 2.2, multiple attempts are made to analyse respiration along with cough and speech signals. Especially, the authors of [10] have reported that the breathing signals are better suited for classifying COVID-19 positive users from healthy users having asthma and a cough. In an another study of [51] with a small dataset of 60 healthy and 20 COVID-19 positive subjects, the authors report better accuracy with RNN based analysis using both breathing and cough data than In an another effort of correlating speech signals with breathing signals, an ensemble system with fusion at both feature and decision level of two approaches is presented in [53] . One of the two approaches is a 1D-CNN 

As such as temperature and demographic information are collected, stored and analysed. During the conversation, the robot monitors the data for cough signals, if not received, it asks the candidate to cough forcefully towards the end of the conversation. At the end, a report is generated and shared with the candidate and authorised doctors.

The increasing count of COVID-19 infectants among healthcare providers suggest the need of telehealth systems where the required care, guidance, and consultation can be provided remotely. The authors of [69] have enlisted the telemedicine providers, regulations, and have discussed the telehealth approach, its benefits, challenges, and the parameters that the clinicians need to look for to confirm the COVID-19 indicators. The major barriers is being the ignorance towards this mode and its usage, trustworthiness of traditional face-to-face communication and preference of interacting with known healthcare providers. The audio-visual assessment that the healthcare providers need to perform includes checking the temperature, ill appearance, calculation of the respiratory rate, presence or absence of cough, and other clinical symptoms. These enlisted parameters can be the focus of the Artificial Intelligence community to bring automation into this space. A bot named "Aapka Chikitsak" [70] was designed using Natural Language Processing and voice-over techniques to cater the needs of the rural Indian population in this time of COVID-19 crisis. It is designed and developed with an intent to provide generic healthcare information, preventive measures, home remedies, and consultation for India-specific context with multi-lingual support to provide healthcare and wellbeing at no extra cost. In addition to the physical diagnostic benefits, [71] talks about the mitigation of psychological problems where psychotherapy sessions can be conducted using video calls.

The need for such a telehealth system is highlighted in [72] , as well as stating the primary advantages such as care for healthcare providers, avoiding overcrowding, and elderly care. However, this also raises the need for proper infrastructure and sufficient bandwidth to enable data transmission (audio, video, and images).

While the readiness of available voice assistants such as Alexa and Siri is analysed in [73] , it is found that they need to be context aware and should be updated with the latest, reliable, and relevant content to be used as a part of a telehealth system for COVID-19. Also, the use of voice as a digital bio marker is missing.

Outside of audio, speech, and language processing, imaging is a dominant signal processing and AI-based diagnosis method. The study performed in [74] on 1 014 cases found that the diagnosis using chest CT has a sensitivity of 97 % in the diagnosis of COVID-19. Multiple efforts by several groups are put in the direction of developing a classifier to detect COVID-19 symptoms using chest X-Ray, such as those in [75] , [76] , [77] , [78] , [79] , and many more. However, as stated in [80] , a multinational consensus is that imaging As concluded by [81] , the chest X-ray findings in COVID-19 patients were found to be peaked at 10-12 days from symptom onset. Also, it is still required to visit a well-equipped clinical facility for such approaches. On a positive note, in the presence of mobile X-Ray machines, this approach can help in speeding up the diagnosis. The authors of [82] have found from an experimental outcome that the chest X-Ray may be useful as a standard method for the rapid diagnosis of COVID-19 to optimise the management of patients.

However, CT is still limited for identifying specific viruses and distinguishing between viruses. The smartphone application framework mentioned in [83] uses multiple sensor data such as that from a camera, microphone, fingerprint sensor, and inertial sensor for feeding CT scan images, human video tracking, as well as capturing cough sounds, temperature, and 30 seconds sit-to-stand movements to the lower layers for processing. With this, the authors plan to detect COVID-19 symptoms such as abnormalities in the CT-Scan images, nausea, fatigue, cough, and fever levels. They further use machine learning techniques such as a CNN and a RNN to derive the COVID-19 result based on the symptoms identified.

In an endeavour to detect the symptoms in each individual before a case gets serious, the government authorities in multiple countries had started examining and asking every individual if they have any COVID-19 symptoms.

Such initiatives need a lot of human efforts to be invested. An alternative automation to accomplish such surveys can be using a system as described in [84] . Using speech recognition, synthesis, and natural language understanding techniques, the CareCall system monitors individuals of Korea and Japan who had a contact with COVID-19 patients. This monitoring is done over a phone call using with and without human-in-the-loop process for three months 

One of the precaution measures while stepping outside home suggested by the WHO to reduce the chance of getting infected or spreading COVID-19

is to wear a facial mask covering nose and mouth. Also, the mathematical analysis presented in [87] suggests that wearing a mask can reduce community Internet Protocol (IP) camera for getting alerts on detecting no-mask on a face. The system provides a facility to add phone numbers for receiving the alerts and also mechanism to see the face not wearing a mask for the admin.

While at quarantine, the doctors need to monitor the cough history of patients, which can be done with a continuous cough monitoring device. After we cross the crisis and organisations think of resuming the operations, continuous monitoring of common spaces such as canteens and lobbies can be realised to record COVID-19 specific coughs. One such monitoring application is developed by the FluSense platform in [89] . It is a surveillance tool to detect influenza like illness from hospital waiting areas using cough sounds. Continuous monitoring and identification of abnormalities from the breathing rate has been done by [90] using image processing. A real-time application of cough detection is also applied using camera devices, which 

The disease spread has equally affected the physical and mental health of the individuals, which is primarily due to plenty of mandatory precautionary measures such as social-distancing, work from home and the quarantine procedures which usually takes around 15 days for the patients to be alone.

Also, the health care providers owing to their hectic routines followed by quarantine days are subject to undergo mental health issues. To cater for the growing need of addressing this issue, not only is there a high demand, but the physical presence of the available psychologists is also missing. As found in [91] , the COVID-19 pandemic has generated unprecedented demand for tele-health based clinical services.

Among several initiatives taken against mental health issues such as stress, anxiety, and depression, we are yet to see these emotions being anal- 

With the smartphone likely being the most convenient and available asset that every individual carries all the time, more of smartphone-based applications for detecting COVID-19 symptoms will help in controlling the virus spread. The authors of [101] have addressed the memory and power consumption issues for importing a deep learning model for detection of cold from the speech signal. They propose network pruning and quantisation techniques to reduce the model size, with which they achieved a size reduction of 95 % in MBytes without affecting the recognition performance.

Along with the physical COVID-19 symptoms, the behavioural aspects also need greater attention. Since the work culture is moving more towards working-from-home, it will be required to detect some behavioural param- A major challenge given the social distancing norms is getting the relevant and accurate speech data for developing machine learning models. Speech enabled chat-bots can play a significant role in this. There are other challenges as well from the design perspective of chat-bots. The authors of [108] have expressed both positive effects and drawbacks associated with using chatbots for sharing the latest information, encouraging desired health im-pacting behaviours, and reducing the psychological damage caused by fear and isolation. This shows that the design of chat-bots should be well thought of for using them, otherwise, they might have negative impact as well. An optimistic approach in these difficult times has been to work towards safe and secure environment for the post pandemic situation so that the society gains the trust and confidence back. This shows the need of accurate and reliable screening and monitoring measures at public places.

Speech and human audio analysis is found to be predominantly useful for COVID-19 analysis. Several initiatives towards identifying cough sound and distinguishing COVID-19 cough from other illnesses are currently taken. It looks promising that soon a cough sound-based sufficiently accurate COVID-19 detector for several real-world use-cases will be available. Such detectors, when integrated with chat-bots can enhance the screening, diagnosing, and monitoring efforts with reduction in human interventions. Further research is required for breathing and speech signal-based COVID-19 analysis, where it is more important to identify the exact bio-markers. With increasing correlation established between speech and breathing signals, detecting breathing disorders from the speech signals will be useful. In many countries, a second, and even a third wave of COVID-19 infection has been found to occur infecting many more individuals. This suggests for the urgent need of robust 

We would like to thank all researchers, health supporters, and others helping in this crisis. Our hearts are with those affected and their families and friends. We acknowledge funding from the German BMWi by ZIM grant

No. 16KN069402 (KIrun).

Covid-19 and computer audition: An overview on what speech & sound analysis could contribute in the sars-cov-2 corona crisis

An overview on audio, signal, speech, & language processing for covid-19

An automated and unobtrusive system for cough detection

Novel feature extraction method for cough detection using nmf

A comparative study of features for acoustic cough detection using deep architectures

The vu sound corpus: Adding more fine-grained annotations to the freesound database

Robust detection of audio-cough events using local hu moments

Continuous sound collection using smartphones and machine learning to measure cough

Infectious diseases society of america guidelines on the diagnosis of covid-19

Exploring automatic diagnosis of covid-19 from crowdsourced respiratory sound data

Nocturnal cough and snore detection using smartphones in presence of multiple background-noises

Deep neural networks for identifying cough sounds

Towards device-agnostic mobile cough detection with convolutional neural networks

Accurate and privacy preserving cough sensing using a low-cost microphone

Automatic cough detection in acoustic signal using spectral features

Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)

A cough-based algorithm for automatic diagnosis of pertussis

Efficient computation of image moments for robust cough detection using smartphones

The coughvid crowdsourcing dataset: A corpus for the study of large-scale cough analysis algorithms

Coswara -a database of breathing, cough, and voice sounds for covid-19 diagnosis

Audio set: An ontology and humanlabeled dataset for audio events

High accuracy classification of covid-19 coughs using mel-frequency cepstral coefficients

neural network with a use case for smart home devices

A novel deep learning based recognition method and web-app for covid-19 infection test from cough sounds with a clinically validated dataset

Cough against covid: Evidence of covid-19 signature in cough sounds

Sars-cov-2 detection from voice

Analysis of acoustic features for speech sound based classification of asthmatic and healthy subjects

The interspeech 2013 computational paralinguistics challenge: Social signals, conflict, emotion, autism

Disease detection using analysis of voice parameters

Automatic wheezing detection using speech recognition technique

Speech signal analysis as an alternative to spirometry in asthma diagnosis: investigating the linear and polynomial correlation coefficient

Assessment of chronic pulmonary disease patients using biomarkers from natural speech recorded by mobile devices

nosis for covid-19 from cough samples via an app

Esc: Dataset for environmental sound classification

Covid-19 artificial intelligence diagnosis using only cough recordings

A review of coronavirus disease-2019 (covid-19)

A covid-19 multipurpose platform

A headset like wearable device to track covid-19 symptoms

Cough classification for covid-19 based on audio mfcc features using convolutional neural networks

Signal analysis and classification of audio samples from individuals diagnosed with covid-19

Studying the similarity of covid-19 sounds based on correlation analysis of mfcc

Speech-based parameter estimation of an asymmetric vocal fold oscillation model and its application in discriminating vocal fold pathologies

Interpreting glottal flow dynamics for detecting covid-19 from voice

Covid-19 patient detection from telephone quality speech data

Audio augmentation for speech recognition

Pay attention to the cough: Early diagnosis of covid-19 using interpretable symptoms embeddings with cough sound signal processing

Endto-end ai-based point-of-care diagnosis system for classifying res

piratory illnesses and early detection of covid-19

Automatic measurement of speech breathing rate

Deep sensing of breathing signal during conversational speech

Smartphone based human breath analysis from respiratory sounds

Human stress detection from the speech in danger situation

Opensmile: the munich versatile and fast open-source audio feature extractor

Covid-19 detection system using recurrent neural networks

The interspeech 2020 computational paralinguistics challenge: Elderly emotion, breathing & masks, Proceedings INTERSPEECH

Ensembling end-to-end deep models for computational paralinguistics tasks: Compare 2020 mask and breathing sub-challenges

Analyzing breath signals for the interspeech 2020 compare challenge

Deep attentive end-to-end continuous breath sensing from speech

Speech breathing estimation using deep learning methods

Obstructive sleep apnea (osa) classification using analysis of breathing sounds during speech

Validation study of watchpat 200 for diagnosis of osa in an asian cohort

Speech as a biomarker for obstructive sleep apnea detection

Reviewing the connection between speech and obstructive sleep apnea

A review of obstructive sleep apnea detection approaches

Can machine learning assist locating the excitation of snore sound? a review

Smartphone-based self-testing of covid-19 using breathing sounds

The remote analysis of breath sound in covid-19 patients: A series of clinical cases, medRxiv

Breath sounds as a biomarker for screening infectious lung diseases

Rale repository of respiratory sounds

An artificial intelligence-based first-line defence against covid-19: digitally screening citizens for risks via a chatbot

A real-time robot-based auxiliary system for risk evaluation of covid-19 infection

Telemedicine in the era of covid-19

Medbot: Conversational artificial intelligence powered chatbot for delivering tele-health after covid-19

on Communication and Electronics Systems (ICCES)

New development:'healing at a distance' -telemedicine and covid-19

New ways to manage pandemics: Using technologies in the era of covid-19, a narrative review

Readiness for voice assistants to support healthcare delivery during a health crisis and pandemic, npj

Correlation of chest ct and rt-pcr testing in coronavirus disease 2019 (covid-19) in china: a report of 1014 cases

Convolutional capsnet: A novel artificial neural network approach to detect covid-19 disease from x-ray images using capsule networks

Covid-caps: A capsule network-based framework for identification of covid-19 cases from x-ray images

Covid-19: automatic detection from x-ray images utilizing transfer learning with convolutional neural networks

Coronavirus (covid-19) classification using ct images by machine learning methods

New machine learning method for image-based diagnosis of covid-19

The role of chest imaging in patient management during the covid-19 pandemic: a multinational consensus statement from the fleischner society

Frequency and distribution of chest radiographic findings in covid-19 positive patients

Coronavirus disease 2019 (covid-19): role of chest ct in diagnosis and management

A novel ai-enabled framework to diagnose coronavirus covid 19 using smartphone embedded sensors: Design study

Carecall: a call-based active monitoring dialog agent for managing covid-19 pandemic

Learning higher representations from pre-trained deep models with data augmentation for the compare 2020 challenge mask task

Paralinguistic classification of mask wearing by image classifiers and fusion

To mask or not to mask: Modeling the potential for face mask use by the general public to curtail the covid-19 pandemic

Masked face recognition dataset and application

Flusense: a contactless syndromic surveillance platform for influenzalike illness in hospital waiting areas

Abnormal respiratory patterns classifier may contribute to large-scale screening of people infected with covid-19 in an accurate and unobtrusive manner

Rapid development of telehealth capabilities within pediatric patient portal infrastructure for covid-19 care: Barriers, solutions, results

An early study on intelligent analysis of speech under covid-19: Severity, sleep quality, fatigue, and anxiety

A mini review on current clinical and research findings for children suffering from covid-19, medRxiv

Is everything fine, grandma? acoustic and linguistic modeling for robust elderly speech emotion recognition

Quantifying the effects of covid-19 on mental health support forums

Dangerous messages or satire? analysing the conspiracy theory linking 5g to covid-19 through social network analysis

Using ai to understand the patient voice during the covid-19 pandemic

Study of coronavirus impact on parisian population from april to june using twitter and text mining approach, medRxiv

How loneliness is talked about in social media during covid-19 pandemic: text mining of 4,492 twitter feeds

Cord-19: The covid-19 open research dataset

Squeeze for sneeze: Compact neural networks for cold and flu recognition, Proceedings INTER-SPEECH

The interspeech 2019 computational paralinguistics challenge: Styrian dialects, continuous sleepiness, baby sounds & orca activity

Using fisher vector and bag-of-audio-words representations to identify styrian dialects, sleepiness, baby & orca sounds, Proceedings of the INTERSPEECH

A novel fusion of attention and sequence to sequence autoencoders to predict sleepiness from speech

Proceedings of the 7th Annual Workshop on Audio/Visual Emotion Challenge

The distress analysis interview corpus of human and computer interviews

Hierarchical attention transfer networks for depression assessment from speech

Chatbots in the fight against the covid-19 pandemic