No | REVISTA | ARTÍCULO | INDEXADA | ISBN/ISSN | FECHA | DOI/ URL |
1 | ICMLT | On the use of VGGish as feature extractor for COVID-19 cough classification. Christian Salamea Palacios, Tarquino Sánchez, Xavier Calderón, Javier Guaña Moya, Paulo Castañeda y Jessica Reina | ACM | 978-1-4503-9832-9 | Jun-23 | https://doi.org/10.1145/3589883.3589896 |
| ABSTRACT | The COVID-19 pandemic has changed the daily lives of all people worldwide, affecting not only society but also various sectors such as finance, tourism, etc. To counteract the pandemic, measures are required to detect contagions and take the necessary actions to prevent the virus spread. In this work, a Transfer Learning approach has been used to model COVID-19 coughs as a previous step to the diagnosis of the illness. The data set of the University of Cambridge, ComParE 2021 COVID-19 Cough Sub-Challenge, has been used, which consists of 725 samples of cough sounds from 397 people of which 119 have been diagnosed with positive COVID-19, besides, a data augmentation technique has been used to balance the data set. This work evaluates the performance of the pre-trained VGGish model for the classification of the audio cough signals as COVID or Not COVID cough. For this purpose, the VGGish model is used as a feature extractor and a convolutional neural network provides the final classification of the cough recordings to determine whether they are COVID-19 positive or negative. Despite the difficulty of the task, optimum results have been founded to detect negative cases obtaining up to 81% of precision. Considering the Unweighted Average Recall (UAR) as metric, the methodology proposed in this work has obtained an improvement up to 3% comparing to OpenSmile technique when the same database has been used. | | | | |
1 | LACCEI | Modelos de aprendizaje automático para caracterizar la señal de la tos de pacientes con COVID-19. Christian Salamea Palacios, Tarquino Sánchez, Xavier Calderón, Javier Guaña Moya, Paulo Castañeda y Jessica Reina | SCOPUS | 978-628-95207-0-5/ 2414-6390 | Jul-22 | http://dx.doi.org/10.18687/LACCEI2022.1.1.145 |
| ABSTRACT | Cough Automatic recognition of audio signals is a challenging signal task due to the difficulty of extracting important attributes from such signals, which relies heavily on discriminating acoustic features to determine the type of cough audio coming from COVID-19 patients. In this work, the use of state-of-the-art pre-trained models and a convolutional neural network for the extraction of characteristics of a cough signal from patients with COVID-19 is analyzed. A comparison of three machine learning models has been proposed to extract the features containing relevant information, leading to the recognition of the COVID-19 cough signal. The first model is based on a basic convolutional neural network, the second is based on a YAMNet pre-treatment model, and the third is a VGGish pre-trained model. The experimental results carried out with a ComPare 2021 CCS database show that models, of the three, used, VGGish to provide better performance when extracting the characteristics of the audio signals of the COVID-19 cough signal, having as results the performance metrics f1 score and accuracy with values of 30.76% and 80.51%, representing an improvement of 6.06% and 3.61% compared to the YANMet model, and the confusion matrices, which validate the mentioned model. | | | | |
2 | SPRINGER | Cough sound identification: an approach based on ensemble learning. Christian Salamea Palacios, Javier Guaña Moya, Tarquino Sanchez, Xavier Calderón and David Naranjo | SCOPUS | 978-981-16-9268-0 | Mar-22 | https://doi.org/10.1007/978-981-16-9268-0_22 |
| ABSTRACT | Cough identification using DSP techniques in an audio signal is a complex task, thus, an artificial intelligence approach is proposed by applying machine learning, deep learning and HMMs algorithms. Later, an ensemble learning model has been used to differentiate cough from other environmental sounds, putting those algorithms together and choosing the best result as the performance of the system. The final system consists of a preprocessing stage where the audio signals are adjusted through data augmentation, normalization, removal of silent fragments, and the transformation to Mel spectrograms. While, on Back-End stage, three models have been evaluated: a convolutional neural network, a random forest, and a classifier based on Hidden Markov Models. We assembled a hard voting classifier (VC) model from the three mod-els to obtain a more robust and reliable model. The VC model reached the high-est precision and F1-score values without False Negative and up to 75% of True Positive values. | | | | |