Search

Your search keyword '"Bredin A"' showing total 4,192 results
Start Over Author "Bredin A"
4,192 results on '"Bredin A"'

1. On the calibration of powerset speaker diarization models

Author

Plaquet, Alexis and Bredin, Hervé

Subjects
Computer Science - Sound, Computer Science - Machine Learning, Electrical Engineering and Systems Science - Audio and Speech Processing
Abstract

End-to-end neural diarization models have usually relied on a multilabel-classification formulation of the speaker diarization problem. Recently, we proposed a powerset multiclass formulation that has beaten the state-of-the-art on multiple datasets. In this paper, we propose to study the calibration of a powerset speaker diarization model, and explore some of its uses. We study the calibration in-domain, as well as out-of-domain, and explore the data in low-confidence regions. The reliability of model confidence is then tested in practice: we use the confidence of the pretrained model to selectively create training and validation subsets out of unannotated data, and compare this to random selection. We find that top-label confidence can be used to reliably predict high-error regions. Moreover, training on low-confidence regions provides a better calibrated model, and validating on low-confidence regions can be more annotation-efficient than random regions.

Published
2024
Full Text
View/download PDF

2. TalTech-IRIT-LIS Speaker and Language Diarization Systems for DISPLACE 2024

Author

Kalda, Joonas, Alumäe, Tanel, Lebourdais, Martin, Bredin, Hervé, Baroudi, Séverin, and Marxer, Ricard

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing
Abstract

This paper describes the submissions of team TalTech-IRIT-LIS to the DISPLACE 2024 challenge. Our team participated in the speaker diarization and language diarization tracks of the challenge. In the speaker diarization track, our best submission was an ensemble of systems based on the pyannote.audio speaker diarization pipeline utilizing powerset training and our recently proposed PixIT method that performs joint diarization and speech separation. We improve upon PixIT by using the separation outputs for speaker embedding extraction. Our ensemble achieved a diarization error rate of 27.1% on the evaluation dataset. In the language diarization track, we fine-tuned a pre-trained Wav2Vec2-BERT language embedding model on in-domain data, and clustered short segments using AHC and VBx, based on similarity scores from LDA/PLDA. This led to a language diarization error rate of 27.6% on the evaluation data. Both results were ranked first in their respective challenge tracks., Comment: accepted at Interspeech 2024

Published
2024

3. PixIT: Joint Training of Speaker Diarization and Speech Separation from Real-world Multi-speaker Recordings

Author

Kalda, Joonas, Pagés, Clément, Marxer, Ricard, Alumäe, Tanel, and Bredin, Hervé

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing
Abstract

A major drawback of supervised speech separation (SSep) systems is their reliance on synthetic data, leading to poor real-world generalization. Mixture invariant training (MixIT) was proposed as an unsupervised alternative that uses real recordings, yet struggles with overseparation and adapting to long-form audio. We introduce PixIT, a joint approach that combines permutation invariant training (PIT) for speaker diarization (SD) and MixIT for SSep. With a small extra requirement of needing SD labels, it solves the problem of overseparation and allows stitching local separated sources leveraging existing work on clustering-based neural SD. We measure the quality of the separated sources via applying automatic speech recognition (ASR) systems to them. PixIT boosts the performance of various ASR systems across two meeting corpora both in terms of the speaker-attributed and utterance-based word error rates while not requiring any fine-tuning., Comment: submitted to Speaker Odyssey 2024

Published
2024
Full Text
View/download PDF

5. Powerset multi-class cross entropy loss for neural speaker diarization

Author

Plaquet, Alexis and Bredin, Hervé

Subjects
Computer Science - Sound, Computer Science - Artificial Intelligence, Computer Science - Computation and Language, Computer Science - Neural and Evolutionary Computing, Electrical Engineering and Systems Science - Audio and Speech Processing
Abstract

Since its introduction in 2019, the whole end-to-end neural diarization (EEND) line of work has been addressing speaker diarization as a frame-wise multi-label classification problem with permutation-invariant training. Despite EEND showing great promise, a few recent works took a step back and studied the possible combination of (local) supervised EEND diarization with (global) unsupervised clustering. Yet, these hybrid contributions did not question the original multi-label formulation. We propose to switch from multi-label (where any two speakers can be active at the same time) to powerset multi-class classification (where dedicated classes are assigned to pairs of overlapping speakers). Through extensive experiments on 9 different benchmarks, we show that this formulation leads to significantly better performance (mostly on overlapping speech) and robustness to domain mismatch, while eliminating the detection threshold hyperparameter, critical for the multi-label formulation.

Published
2023
Full Text
View/download PDF

6. The biogeography of the Amazonian tree flora

Author

Bruno Garcia Luize, Hanna Tuomisto, Robin Ekelschot, Kyle G. Dexter, Iêda L. do Amaral, Luiz de Souza Coelho, Francisca Dionízia de Almeida Matos, Diógenes de Andrade Lima Filho, Rafael P. Salomão, Florian Wittmann, Carolina V. Castilho, Marcelo de Jesus Veiga Carim, Juan Ernesto Guevara, Oliver L. Phillips, William E. Magnusson, Daniel Sabatier, Juan David Cardenas Revilla, Jean-François Molino, Mariana Victória Irume, Maria Pires Martins, José Renan da Silva Guimarães, José Ferreira Ramos, Olaf S. Bánki, Maria Teresa Fernandez Piedade, Dairon Cárdenas López, Nigel C. A. Pitman, Layon O. Demarchi, Jochen Schöngart, Evlyn Márcia Moraes de Leão Novo, Percy Núñez Vargas, Thiago Sanna Freire Silva, Eduardo Martins Venticinque, Angelo Gilberto Manzatto, Neidiane Farias Costa Reis, John Terborgh, Katia Regina Casula, Euridice N. Honorio Coronado, Abel Monteagudo Mendoza, Juan Carlos Montero, Flávia R. C. Costa, Ted R. Feldpausch, Adriano Costa Quaresma, Nicolás Castaño Arboleda, Charles Eugene Zartman, Timothy J. Killeen, Beatriz S. Marimon, Ben Hur Marimon, Rodolfo Vasquez, Bonifacio Mostacedo, Rafael L. Assis, Chris Baraloto, Dário Dantas do Amaral, Julien Engel, Pascal Petronelli, Hernán Castellanos, Marcelo Brilhante de Medeiros, Marcelo Fragomeni Simon, Ana Andrade, José Luís Camargo, William F. Laurance, Susan G. W. Laurance, Lorena Maniguaje Rincón, Juliana Schietti, Thaiane R. Sousa, Gisele Biem Mori, Emanuelle de Sousa Farias, Maria Aparecida Lopes, José Leonardo Lima Magalhães, Henrique Eduardo Mendonça Nascimento, Helder Lima de Queiroz, Caroline C. Vasconcelos, Gerardo A. Aymard C, Roel Brienen, Pablo R. Stevenson, Alejandro Araujo-Murakami, Bruno Barçante Ladvocat Cintra, Tim R. Baker, Yuri Oliveira Feitosa, Hugo F. Mogollón, Joost F. Duivenvoorden, Carlos A. Peres, Miles R. Silman, Leandro Valle Ferreira, José Rafael Lozada, James A. Comiskey, José Julio de Toledo, Gabriel Damasco, Nállarett Dávila, Freddie C. Draper, Roosevelt García-Villacorta, Aline Lopes, Alberto Vicentini, Fernando Cornejo Valverde, Alfonso Alonso, Luzmila Arroyo, Francisco Dallmeier, Vitor H. F. Gomes, Eliana M. Jimenez, David Neill, Maria Cristina Peñuela Mora, Janaína Costa Noronha, Daniel P. P. de Aguiar, Flávia Rodrigues Barbosa, Yennie K. Bredin, Rainiellen de Sá Carpanedo, Fernanda Antunes Carvalho, Fernanda Coelho de Souza, Kenneth J. Feeley, Rogerio Gribel, Torbjørn Haugaasen, Joseph E. Hawes, Marcelo Petratti Pansonato, John J. Pipoly, Marcos Ríos Paredes, Domingos de Jesus Rodrigues, Jos Barlow, Erika Berenguer, Izaias Brasil da Silva, Maria Julia Ferreira, Joice Ferreira, Paul V. A. Fine, Marcelino Carneiro Guedes, Carolina Levis, Juan Carlos Licona, Boris Eduardo Villa Zegarra, Vincent Antoine Vos, Carlos Cerón, Flávia Machado Durgante, Émile Fonty, Terry W. Henkel, John Ethan Householder, Isau Huamantupa-Chuquimaco, Marcos Silveira, Juliana Stropp, Raquel Thomas, Doug Daly, William Milliken, Guido Pardo Molina, Toby Pennington, Ima Célia Guimarães Vieira, Bianca Weiss Albuquerque, Wegliane Campelo, Alfredo Fuentes, Bente Klitgaard, José Luis Marcelo Pena, J. Sebastián Tello, Corine Vriesendorp, Jerome Chave, Anthony Di Fiore, Renato Richard Hilário, Luciana de Oliveira Pereira, Juan Fernando Phillips, Gonzalo Rivas-Torres, Tinde R. van Andel, Patricio von Hildebrand, William Balee, Edelcilio Marques Barbosa, Luiz Carlos de Matos Bonates, Hilda Paulette Dávila Doza, Ricardo Zárate Gómez, Therany Gonzales, George Pepe Gallardo Gonzales, Bruce Hoffman, André Braga Junqueira, Yadvinder Malhi, Ires Paula de Andrade Miranda, Linder Felipe Mozombite Pinto, Adriana Prieto, Agustín Rudas, Ademir R. Ruschel, Natalino Silva, César I. A. Vela, Stanford Zent, Egleé L. Zent, María José Endara, Angela Cano, Yrma Andreina Carrero Márquez, Diego F. Correa, Janaina Barbosa Pedrosa Costa, Bernardo Monteiro Flores, David Galbraith, Milena Holmgren, Michelle Kalamandeen, Guilherme Lobo, Luis Torres Montenegro, Marcelo Trindade Nascimento, Alexandre A. Oliveira, Maihyra Marina Pombo, Hirma Ramirez-Angulo, Maira Rocha, Veridiana Vizoni Scudeller, Maria Natalia Umaña, Geertje van der Heijden, Emilio Vilanova Torre, Tony Mori Vargas, Manuel Augusto Ahuite Reategui, Cláudia Baider, Henrik Balslev, Sasha Cárdenas, Luisa Fernanda Casas, William Farfan-Rios, Cid Ferreira, Reynaldo Linares-Palomino, Casimiro Mendoza, Italo Mesones, Germaine Alexander Parada, Armando Torres-Lezama, Ligia Estela Urrego Giraldo, Daniel Villarroel, Roderick Zagt, Miguel N. Alexiades, Edmar Almeida de Oliveira, Riley P. Fortier, Karina Garcia-Cabrera, Lionel Hernandez, Walter Palacios Cuenca, Susamar Pansini, Daniela Pauletto, Freddy Ramirez Arevalo, Adeilza Felipe Sampaio, Elvis H. Valderrama Sandoval, Luis Valenzuela Gamarra, Marina Hirota, Clarisse Palma-Silva, and Hans ter Steege

Subjects
Biology (General), QH301-705.5
Abstract

Abstract We describe the geographical variation in tree species composition across Amazonian forests and show how environmental conditions are associated with species turnover. Our analyses are based on 2023 forest inventory plots (1 ha) that provide abundance data for a total of 5188 tree species. Within-plot species composition reflected both local environmental conditions (especially soil nutrients and hydrology) and geographical regions. A broader-scale view of species turnover was obtained by interpolating the relative tree species abundances over Amazonia into 47,441 0.1-degree grid cells. Two main dimensions of spatial change in tree species composition were identified. The first was a gradient between western Amazonia at the Andean forelands (with young geology and relatively nutrient-rich soils) and central–eastern Amazonia associated with the Guiana and Brazilian Shields (with more ancient geology and poor soils). The second gradient was between the wet forests of the northwest and the drier forests in southern Amazonia. Isolines linking cells of similar composition crossed major Amazonian rivers, suggesting that tree species distributions are not limited by rivers. Even though some areas of relatively sharp species turnover were identified, mostly the tree species composition changed gradually over large extents, which does not support delimiting clear discrete biogeographic regions within Amazonia.

Published
2024
Full Text
View/download PDF

7. Impact of critical illness on continuation of anticancer treatment and prognosis of patients with aggressive hematological malignancies

Author

Swann Bredin, Justine Decroocq, Clément Devautour, Julien Charpentier, Clara Vigneron, and Frédéric Pène

Subjects
Leukemia, Lymphoma, Chemotherapy, Outcome, ICU, Medical emergencies. Critical care. Intensive care. First aid, RC86-88.9
Abstract

Abstract Background Maintaining the dose-intensity of cancer treatment is an important prognostic factor of aggressive hematological malignancies. The objective of this study was to assess the long-term outcomes of intensive care unit (ICU) survivors with acute myeloid leukemia (AML) or aggressive B-cell non-Hodgkin lymphoma (B-NHL) with emphasis on the resumption of the intended optimal regimen of cancer treatment. Patients and methods We conducted a retrospective (2013–2021) single-center observational study where we included patients with AML and B-NHL discharged alive from the ICU after an unplanned admission. The primary endpoint was the change in the intended optimal cancer treatment following ICU discharge. Secondary endpoints were 1-year progression-free survival and overall survival rates. Determinants associated with modifications in cancer treatment were assessed through multivariate logistic regression. Results Over the study period, 366 patients with AML or B-NHL were admitted to the ICU, of whom 170 survivors with AML (n = 92) and B-NHL (n = 78) formed the cohort of interest. The hematological malignancy was recently diagnosed in 68% of patients. The admission Sequential Organ Failure Assessment (SOFA) score was 5 (interquartile range 4–8). During the ICU stay, 30 patients (17.6%) required invasive mechanical ventilation, 29 (17.0%) vasopressor support, and 16 (9.4%) renal replacement therapy. The one-year survival rate following ICU discharge was 59.5%. Further modifications in hematologic treatment regimens were required in 72 patients (42%). In multivariate analysis, age > 65 years (odds ratio (OR) 3.54 [95%-confidence interval 1.67–7.50], p 20 µmol/L (OR 3.01 [1.10–8.15], p = 0.031), and therapeutic limitations (OR 16.5 [1.83–149.7], p = 0.012) were independently associated with modifications in cancer treatment. Post-ICU modifications of cancer treatment had significant impact on in-hospital, 1-year overall survival and progression-free survival. Conclusion The intended cancer treatment could be resumed in 58% of ICU survivors with aggressive hematological malignancies. At the time of ICU discharge, advanced age, persistent liver dysfunction and decisions to limit further life-support therapies were independent determinants of cancer treatment modifications. These modifications were associated with worsened one-year outcomes.

Published
2024
Full Text
View/download PDF

8. Language Skills Influence Transition Planning in Adolescents with Fragile X Syndrome

Author

Olivia Boorom, Heather Fielding-Gebhardt, Shelley Bredin-Oja, Kandace Fleming, Rebecca E. Swinburne Romine, and Nancy Brady

Abstract

Individuals with fragile X syndrome (FXS) and their parents have a range of experiences navigating the crucial transition period between adolescence and adulthood. Semi structured interviews of 47 mothers of adolescents with FXS (mean child age = 15.89 years) were analyzed to identify mothers' changing expectations during the adolescent period and parent goals related to work and postsecondary education. Mothers' work and education goals were explored in relation to child factors such as language skills and autism characteristics. Lower language skills were associated with lower likelihood of reporting vocational goals. Results suggest that adolescents with FXS with lower language ability are less likely to have vocational plans for adulthood during this critical period and may need greater transition planning assistance.

Published
2024

9. BabySLM: language-acquisition-friendly benchmark of self-supervised spoken language models

Author

Lavechin, Marvin, Sy, Yaya, Titeux, Hadrien, Blandón, María Andrea Cruz, Räsänen, Okko, Bredin, Hervé, Dupoux, Emmanuel, and Cristia, Alejandrina

Subjects
Computer Science - Computation and Language, Electrical Engineering and Systems Science - Audio and Speech Processing, Statistics - Machine Learning
Abstract

Self-supervised techniques for learning speech representations have been shown to develop linguistic competence from exposure to speech without the need for human labels. In order to fully realize the potential of these approaches and further our understanding of how infants learn language, simulations must closely emulate real-life situations by training on developmentally plausible corpora and benchmarking against appropriate test sets. To this end, we propose a language-acquisition-friendly benchmark to probe spoken language models at the lexical and syntactic levels, both of which are compatible with the vocabulary typical of children's language experiences. This paper introduces the benchmark and summarizes a range of experiments showing its usefulness. In addition, we highlight two exciting challenges that need to be addressed for further progress: bridging the gap between text and speech and between clean speech and in-the-wild speech., Comment: Proceedings of Interspeech 2023

Published
2023

10. The biogeography of the Amazonian tree flora

Author

Luize, Bruno Garcia, Tuomisto, Hanna, Ekelschot, Robin, Dexter, Kyle G., Amaral, Iêda L. do, Coelho, Luiz de Souza, Matos, Francisca Dionízia de Almeida, Lima Filho, Diógenes de Andrade, Salomão, Rafael P., Wittmann, Florian, Castilho, Carolina V., Carim, Marcelo de Jesus Veiga, Guevara, Juan Ernesto, Phillips, Oliver L., Magnusson, William E., Sabatier, Daniel, Cardenas Revilla, Juan David, Molino, Jean-François, Irume, Mariana Victória, Martins, Maria Pires, Guimarães, José Renan da Silva, Ramos, José Ferreira, Bánki, Olaf S., Piedade, Maria Teresa Fernandez, Cárdenas López, Dairon, Pitman, Nigel C. A., Demarchi, Layon O., Schöngart, Jochen, de Leão Novo, Evlyn Márcia Moraes, Núñez Vargas, Percy, Silva, Thiago Sanna Freire, Venticinque, Eduardo Martins, Manzatto, Angelo Gilberto, Reis, Neidiane Farias Costa, Terborgh, John, Casula, Katia Regina, Honorio Coronado, Euridice N., Mendoza, Abel Monteagudo, Montero, Juan Carlos, Costa, Flávia R. C., Feldpausch, Ted R., Quaresma, Adriano Costa, Castaño Arboleda, Nicolás, Zartman, Charles Eugene, Killeen, Timothy J., Marimon, Beatriz S., Marimon, Ben Hur, Vasquez, Rodolfo, Mostacedo, Bonifacio, Assis, Rafael L., Baraloto, Chris, do Amaral, Dário Dantas, Engel, Julien, Petronelli, Pascal, Castellanos, Hernán, de Medeiros, Marcelo Brilhante, Simon, Marcelo Fragomeni, Andrade, Ana, Camargo, José Luís, Laurance, William F., Laurance, Susan G. W., Rincón, Lorena Maniguaje, Schietti, Juliana, Sousa, Thaiane R., Mori, Gisele Biem, Farias, Emanuelle de Sousa, Lopes, Maria Aparecida, Magalhães, José Leonardo Lima, Nascimento, Henrique Eduardo Mendonça, de Queiroz, Helder Lima, Vasconcelos, Caroline C., Aymard C, Gerardo A., Brienen, Roel, Stevenson, Pablo R., Araujo-Murakami, Alejandro, Cintra, Bruno Barçante Ladvocat, Baker, Tim R., Feitosa, Yuri Oliveira, Mogollón, Hugo F., Duivenvoorden, Joost F., Peres, Carlos A., Silman, Miles R., Ferreira, Leandro Valle, Lozada, José Rafael, Comiskey, James A., de Toledo, José Julio, Damasco, Gabriel, Dávila, Nállarett, Draper, Freddie C., García-Villacorta, Roosevelt, Lopes, Aline, Vicentini, Alberto, Valverde, Fernando Cornejo, Alonso, Alfonso, Arroyo, Luzmila, Dallmeier, Francisco, Gomes, Vitor H. F., Jimenez, Eliana M., Neill, David, Peñuela Mora, Maria Cristina, Noronha, Janaína Costa, de Aguiar, Daniel P. P., Barbosa, Flávia Rodrigues, Bredin, Yennie K., Carpanedo, Rainiellen de Sá, Carvalho, Fernanda Antunes, Souza, Fernanda Coelho de, Feeley, Kenneth J., Gribel, Rogerio, Haugaasen, Torbjørn, Hawes, Joseph E., Pansonato, Marcelo Petratti, Pipoly, III, John J., Paredes, Marcos Ríos, Rodrigues, Domingos de Jesus, Barlow, Jos, Berenguer, Erika, da Silva, Izaias Brasil, Ferreira, Maria Julia, Ferreira, Joice, Fine, Paul V. A., Guedes, Marcelino Carneiro, Levis, Carolina, Licona, Juan Carlos, Villa Zegarra, Boris Eduardo, Vos, Vincent Antoine, Cerón, Carlos, Durgante, Flávia Machado, Fonty, Émile, Henkel, Terry W., Householder, John Ethan, Huamantupa-Chuquimaco, Isau, Silveira, Marcos, Stropp, Juliana, Thomas, Raquel, Daly, Doug, Milliken, William, Molina, Guido Pardo, Pennington, Toby, Vieira, Ima Célia Guimarães, Albuquerque, Bianca Weiss, Campelo, Wegliane, Fuentes, Alfredo, Klitgaard, Bente, Pena, José Luis Marcelo, Tello, J. Sebastián, Vriesendorp, Corine, Chave, Jerome, Di Fiore, Anthony, Hilário, Renato Richard, Pereira, Luciana de Oliveira, Phillips, Juan Fernando, Rivas-Torres, Gonzalo, van Andel, Tinde R., von Hildebrand, Patricio, Balee, William, Barbosa, Edelcilio Marques, Bonates, Luiz Carlos de Matos, Dávila Doza, Hilda Paulette, Zárate Gómez, Ricardo, Gonzales, Therany, Gallardo Gonzales, George Pepe, Hoffman, Bruce, Junqueira, André Braga, Malhi, Yadvinder, Miranda, Ires Paula de Andrade, Pinto, Linder Felipe Mozombite, Prieto, Adriana, Rudas, Agustín, Ruschel, Ademir R., Silva, Natalino, Vela, César I. A., Zent, Stanford, Zent, Egleé L., Endara, María José, Cano, Angela, Carrero Márquez, Yrma Andreina, Correa, Diego F., Costa, Janaina Barbosa Pedrosa, Monteiro Flores, Bernardo, Galbraith, David, Holmgren, Milena, Kalamandeen, Michelle, Lobo, Guilherme, Torres Montenegro, Luis, Nascimento, Marcelo Trindade, Oliveira, Alexandre A., Pombo, Maihyra Marina, Ramirez-Angulo, Hirma, Rocha, Maira, Scudeller, Veridiana Vizoni, Umaña, Maria Natalia, van der Heijden, Geertje, Vilanova Torre, Emilio, Vargas, Tony Mori, Ahuite Reategui, Manuel Augusto, Baider, Cláudia, Balslev, Henrik, Cárdenas, Sasha, Casas, Luisa Fernanda, Farfan-Rios, William, Ferreira, Cid, Linares-Palomino, Reynaldo, Mendoza, Casimiro, Mesones, Italo, Parada, Germaine Alexander, Torres-Lezama, Armando, Urrego Giraldo, Ligia Estela, Villarroel, Daniel, Zagt, Roderick, Alexiades, Miguel N., de Oliveira, Edmar Almeida, Fortier, Riley P., Garcia-Cabrera, Karina, Hernandez, Lionel, Palacios Cuenca, Walter, Pansini, Susamar, Pauletto, Daniela, Ramirez Arevalo, Freddy, Sampaio, Adeilza Felipe, Valderrama Sandoval, Elvis H., Valenzuela Gamarra, Luis, Hirota, Marina, Palma-Silva, Clarisse, and ter Steege, Hans

Published
2024
Full Text
View/download PDF

15. Bloodstream infections among patients receiving therapeutic plasma exchanges in the intensive care unit: a 10 year multicentric study

Author

Sofiane Fodil, Tomas Urbina, Swann Bredin, Julien Mayaux, Antoine Lafarge, Louaï Missri, Eric Maury, Alexandre Demoule, Frederic Pene, Eric Mariotte, and Hafid Ait-Oufella

Subjects
Therapeutic plasma exchange, Intensive care unit, Critical care, Bloodstream infections, Intra-vascular catheter infection, Health-care acquired infections, Medical emergencies. Critical care. Intensive care. First aid, RC86-88.9
Abstract

Abstract Background Therapeutic plasma exchanges (TPE), which affect the humoral response, are often performed in combination with immunosuppressive drugs. For this reason, TPE may be associated with an increased susceptibility to infections. We aimed to describe blood stream infection (BSI) incidence in ICU patients treated with TPE and to identify associated risk factors. Methods We retrospectively included patients that had received at least one session of TPE in the ICU of one of the 4 participating centers (all in Paris, France) between January 1st 2010 and December 31th 2019. Patients presenting with a BSI during ICU stay were compared to patients without such an infection. Risk factors for BSI were identified by a multivariate logistic regression model. Results Over 10 years in the 4 ICUs, 387 patients were included, with a median of 5 [2–7] TPE sessions per patient. Most frequent indications for TPE were thrombotic microangiopathy (47%), central nervous system inflammatory disorders (11%), hyperviscosity syndrome (11%) and ANCA associated vasculitis (8.5%). Thirty-one patients (8%) presented with a BSI during their ICU stay, a median of 7 [3–11] days after start of TPE. In a multivariate logistic regression model, diabetes (OR 3.32 [1.21–8.32]) and total number of TPE sessions (OR 1.14 [1.08–1.20]) were independent risk factors for BSI. There was no difference between TPE catheter infection related BSI (n = 11 (35%)) and other sources of BSI (n = 20 (65%)) regarding catheter insertion site (p = 0.458) or rate of TPE catheter related deep vein thrombosis (p = 0.601). ICU course was severe in patients presenting with BSI when compared to patients without BSI, with higher need for mechanical ventilation (45% vs 18%, p = 0.001), renal replacement therapy (42% vs 20%, p = 0.011), vasopressors (32% vs 12%, p = 0.004) and a higher mortality (19% vs 5%, p = 0.010). Conclusion Blood stream infections are frequent in patients receiving TPE in the ICU, and are associated with a severe ICU course. Vigilant monitoring is crucial particularly for patients receiving a high number of TPE sessions.

Published
2024
Full Text
View/download PDF

16. The relationship between off-ice testing and on-ice performance in male youth Ice hockey players

Author

Mark S. Rice, Darren E. R. Warburton, Alejandro Gaytan-Gonzalez, Veronica K. Jamnik, Kai Kaufman, Declan R. D. Warburton, Michael Souster, and Shannon S. D. Bredin

Subjects
off-ice testing, on-ice performance, ice hockey, youth players, predictive metrics, Sports, GV557-1198.995
Abstract

IntroductionIce hockey demands a unique blend of physical fitness and skill, necessitating a comprehensive understanding of the factors influencing on-ice performance. The present study was designed to examine the relationship between off-ice and on-ice performance measures in male, youth, ice hockey players.MethodsEleven minor hockey players (Age = 9.8 ± 1.1 years) participated in two testing days: (1) off-ice and (2) on-ice assessments. Off-ice assessments included maximal aerobic power, anaerobic fitness, muscular strength (handgrip and single leg squat), muscular endurance (curl-ups and push-ups), muscular power (standing long jump and vertical jump), and 30 m sprinting speed and acceleration. On-ice testing included a 15.2 m maximum speed test, a 6.1 m acceleration test with a continuation into a 47.9 m top speed test, an agility cornering S turn test, and a shot velocity test.ResultsTwenty-four out of 33 off-ice variables were significantly correlated with at least one of the 11 on-ice performance variables. From those 24, 10 were included as predictors for at least one of the on-ice performance variables. Each model was composed of either one or two predictors, where the most common predictors were 30 m Run – Split (6.1 m) Sprint time and 30 m Run – Total (30 m) Sprint Max speed (included in four out of 11 models each). The prediction formulas R2 and coefficient of variation ranged from 0.63% to 0.96% and 1.2% to 15.3%, respectively.DiscussionDiverse off-ice measures of aerobic fitness, anaerobic power, muscular strength, power, and endurance, and sprinting speed, acceleration, and agility are predictive of on-ice performance. The insights gained from this study contribute to the refinement of assessment protocols, fostering a comprehensive approach to optimizing player performance and potential. Understanding the connection between objective off-ice testing and on-ice performance can support tailored training programs and player development in male youth ice hockey.

Published
2024
Full Text
View/download PDF

17. The Seventh Circuit

Author

Larson, William B. and Bredin, Wade A.

Subjects
United States. Court of Appeals for the 7th Circuit -- Surveys, Evidence (Law) -- Surveys, Insurance, Law
Abstract

Manpower, Inc. v. Insurance Company of the State of Pennsylvania (1) THE partial collapse of a Paris office building resulted in insurance litigation over the cost of business interruption. The [...]

Published
2024

21. Brouhaha: multi-task training for voice activity detection, speech-to-noise ratio, and C50 room acoustics estimation

Author

Lavechin, Marvin, Métais, Marianne, Titeux, Hadrien, Boissonnet, Alodie, Copet, Jade, Rivière, Morgane, Bergelson, Elika, Cristia, Alejandrina, Dupoux, Emmanuel, and Bredin, Hervé

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Sound
Abstract

Most automatic speech processing systems register degraded performance when applied to noisy or reverberant speech. But how can one tell whether speech is noisy or reverberant? We propose Brouhaha, a neural network jointly trained to extract speech/non-speech segments, speech-to-noise ratios, and C50room acoustics from single-channel recordings. Brouhaha is trained using a data-driven approach in which noisy and reverberant audio segments are synthesized. We first evaluate its performance and demonstrate that the proposed multi-task regime is beneficial. We then present two scenarios illustrating how Brouhaha can be used on naturally noisy and reverberant data: 1) to investigate the errors made by a speaker diarization model (pyannote.audio); and 2) to assess the reliability of an automatic speech recognition model (Whisper from OpenAI). Both our pipeline and a pretrained model are open source and shared with the speech community.

Published
2022

23. Neoadjuvant trastuzumab deruxtecan (T-DXd) with response-directed definitive therapy in early stage HER2-positive breast cancer: a phase II study protocol (SHAMROCK study)

Author

Gavin P. Dowling, Sinead Toomey, Philip Bredin, Imelda Parker, Eibhlin Mulroe, Jacinta Marron, Olivia McLoughlin, Ausra Teiserskiene, Colm Power, Anne Marie O’Shea, Megan Greally, Patrick G. Morris, Deirdre Duke, Arnold D. K. Hill, and Bryan T. Hennessy

Subjects
Breast cancer, Trastuzumab deruxtecan, HER2-positive, Neoadjuvant, Pathological Complete Response, Neoplasms. Tumors. Oncology. Including cancer and carcinogens, RC254-282
Abstract

Abstract Background The current standard of care in the neoadjuvant setting for high-risk HER2-positive (HER2 +) breast cancer is to combine systemic chemotherapy with dual HER2 blockade, trastuzumab and pertuzumab. Targeted therapies have significantly improved outcomes for patients with HER2-positive breast cancer. To improve treatment-associated toxicity, chemotherapy-sparing approaches are currently being investigated. Trastuzumab deruxtecan (T-DXd) is an HER2-directed antibody–drug-conjugate (ADC) with promising results in the metastatic setting for HER2-positive breast cancer. The SHAMROCK study investigates neoadjuvant T-DXd in early stage HER2-positive breast cancer, using pathological complete response (pCR) rate as the primary endpoint. Methods This is a phase II open-label, single arm, adaptive multi-centre trial of T-DXd in the neoadjuvant setting in stage 2–3 HER2-positive breast cancer. Eligible patients will receive 5.4 mg/kg of T-DXd intravenously every 3 weeks for up to 6 cycles. A repeat biopsy will performed after 2 cycles for the RNA disruption index (RDI) score assessment. According to their likelihood of pCR, as determined by the RDI score, patients will either undergo 4 or 6 cycles of T-DXd prior to imaging. Patients with imaging complete response (iCR) after either 4 or 6 cycles will proceed to surgery. Patients who do not achieve iCR will either undergo further systemic therapy or proceed to surgery. Discussion The SHAMROCK study is a chemotherapy-sparing approach to curative intent treatment, investigating neoadjuvant T-DXd. We hypothesise that neoadjuvant T-DXd will have a high pCR rate and be associated low toxicity in early stage HER2-positive breast cancer. Trial registration EudraCT Number: 2022–002485-32; ClinicalTrials.gov identifier: NCT05710666; Cancer Trials Ireland study number: CTRIAL-IE 22–01.

Published
2024
Full Text
View/download PDF

25. Mapping density, diversity and species-richness of the Amazon tree flora

Author

Hans ter Steege, Nigel C. A. Pitman, Iêda Leão do Amaral, Luiz de Souza Coelho, Francisca Dionízia de Almeida Matos, Diógenes de Andrade Lima Filho, Rafael P. Salomão, Florian Wittmann, Carolina V. Castilho, Juan Ernesto Guevara, Marcelo de Jesus Veiga Carim, Oliver L. Phillips, William E. Magnusson, Daniel Sabatier, Juan David Cardenas Revilla, Jean-François Molino, Mariana Victória Irume, Maria Pires Martins, José Renan da Silva Guimarães, José Ferreira Ramos, Olaf S. Bánki, Maria Teresa Fernandez Piedade, Dairon Cárdenas López, Domingos de Jesus Rodrigues, Layon O. Demarchi, Jochen Schöngart, Everton José Almeida, Luciane Ferreira Barbosa, Larissa Cavalheiro, Márcia Cléia Vilela dos Santos, Bruno Garcia Luize, Evlyn Márcia Moraes de Leão Novo, Percy Núñez Vargas, Thiago Sanna Freire Silva, Eduardo Martins Venticinque, Angelo Gilberto Manzatto, Neidiane Farias Costa Reis, John Terborgh, Katia Regina Casula, Euridice N. Honorio Coronado, Abel Monteagudo Mendoza, Juan Carlos Montero, Flávia R. C. Costa, Ted R. Feldpausch, Adriano Costa Quaresma, Nicolás Castaño Arboleda, Charles Eugene Zartman, Timothy J. Killeen, Beatriz S. Marimon, Ben Hur Marimon-Junior, Rodolfo Vasquez, Bonifacio Mostacedo, Rafael L. Assis, Chris Baraloto, Dário Dantas do Amaral, Julien Engel, Pascal Petronelli, Hernán Castellanos, Marcelo Brilhante de Medeiros, Marcelo Fragomeni Simon, Ana Andrade, José Luís Camargo, William F. Laurance, Susan G. W. Laurance, Lorena Maniguaje Rincón, Juliana Schietti, Thaiane R. Sousa, Emanuelle de Sousa Farias, Maria Aparecida Lopes, José Leonardo Lima Magalhães, Henrique Eduardo Mendonça Nascimento, Helder Lima de Queiroz, Gerardo A. Aymard C., Roel Brienen, Pablo R. Stevenson, Alejandro Araujo-Murakami, Tim R. Baker, Bruno Barçante Ladvocat Cintra, Yuri Oliveira Feitosa, Hugo F. Mogollón, Joost F. Duivenvoorden, Carlos A. Peres, Miles R. Silman, Leandro Valle Ferreira, José Rafael Lozada, James A. Comiskey, Freddie C. Draper, José Julio de Toledo, Gabriel Damasco, Roosevelt García-Villacorta, Aline Lopes, Alberto Vicentini, Fernando Cornejo Valverde, Alfonso Alonso, Luzmila Arroyo, Francisco Dallmeier, Vitor H. F. Gomes, Eliana M. Jimenez, David Neill, Maria Cristina Peñuela Mora, Janaína Costa Noronha, Daniel P. P. de Aguiar, Flávia Rodrigues Barbosa, Yennie K. Bredin, Rainiellen de Sá Carpanedo, Fernanda Antunes Carvalho, Fernanda Coelho de Souza, Kenneth J. Feeley, Rogerio Gribel, Torbjørn Haugaasen, Joseph E. Hawes, Marcelo Petratti Pansonato, Marcos Ríos Paredes, Jos Barlow, Erika Berenguer, Izaias Brasil da Silva, Maria Julia Ferreira, Joice Ferreira, Paul V. A. Fine, Marcelino Carneiro Guedes, Carolina Levis, Juan Carlos Licona, Boris Eduardo Villa Zegarra, Vincent Antoine Vos, Carlos Cerón, Flávia Machado Durgante, Émile Fonty, Terry W. Henkel, John Ethan Householder, Isau Huamantupa-Chuquimaco, Edwin Pos, Marcos Silveira, Juliana Stropp, Raquel Thomas, Doug Daly, Kyle G. Dexter, William Milliken, Guido Pardo Molina, Toby Pennington, Ima Célia Guimarães Vieira, Bianca Weiss Albuquerque, Wegliane Campelo, Alfredo Fuentes, Bente Klitgaard, José Luis Marcelo Pena, J. Sebastián Tello, Corine Vriesendorp, Jerome Chave, Anthony Di Fiore, Renato Richard Hilário, Luciana de Oliveira Pereira, Juan Fernando Phillips, Gonzalo Rivas-Torres, Tinde R. van Andel, Patricio von Hildebrand, William Balee, Edelcilio Marques Barbosa, Luiz Carlos de Matos Bonates, Hilda Paulette Dávila Doza, Ricardo Zárate Gómez, Therany Gonzales, George Pepe Gallardo Gonzales, Bruce Hoffman, André Braga Junqueira, Yadvinder Malhi, Ires Paula de Andrade Miranda, Linder Felipe Mozombite Pinto, Adriana Prieto, Agustín Rudas, Ademir R. Ruschel, Natalino Silva, César I. A. Vela, Egleé L. Zent, Stanford Zent, Angela Cano, Yrma Andreina Carrero Márquez, Diego F. Correa, Janaina Barbosa Pedrosa Costa, Bernardo Monteiro Flores, David Galbraith, Milena Holmgren, Michelle Kalamandeen, Guilherme Lobo, Luis Torres Montenegro, Marcelo Trindade Nascimento, Alexandre A. Oliveira, Maihyra Marina Pombo, Hirma Ramirez-Angulo, Maira Rocha, Veridiana Vizoni Scudeller, Rodrigo Sierra, Milton Tirado, Maria Natalia Umaña, Geertje van der Heijden, Emilio Vilanova Torre, Manuel Augusto Ahuite Reategui, Cláudia Baider, Henrik Balslev, Sasha Cárdenas, Luisa Fernanda Casas, María José Endara, William Farfan-Rios, Cid Ferreira, Reynaldo Linares-Palomino, Casimiro Mendoza, Italo Mesones, Germaine Alexander Parada, Armando Torres-Lezama, Ligia Estela Urrego Giraldo, Daniel Villarroel, Roderick Zagt, Miguel N. Alexiades, Edmar Almeida de Oliveira, Karina Garcia-Cabrera, Lionel Hernandez, Walter Palacios Cuenca, Susamar Pansini, Daniela Pauletto, Freddy Ramirez Arevalo, Adeilza Felipe Sampaio, Elvis H. Valderrama Sandoval, Luis Valenzuela Gamarra, Aurora Levesley, Georgia Pickavance, and Karina Melgaço

Subjects
Biology (General), QH301-705.5
Abstract

Abstract Using 2.046 botanically-inventoried tree plots across the largest tropical forest on Earth, we mapped tree species-diversity and tree species-richness at 0.1-degree resolution, and investigated drivers for diversity and richness. Using only location, stratified by forest type, as predictor, our spatial model, to the best of our knowledge, provides the most accurate map of tree diversity in Amazonia to date, explaining approximately 70% of the tree diversity and species-richness. Large soil-forest combinations determine a significant percentage of the variation in tree species-richness and tree alpha-diversity in Amazonian forest-plots. We suggest that the size and fragmentation of these systems drive their large-scale diversity patterns and hence local diversity. A model not using location but cumulative water deficit, tree density, and temperature seasonality explains 47% of the tree species-richness in the terra-firme forest in Amazonia. Over large areas across Amazonia, residuals of this relationship are small and poorly spatially structured, suggesting that much of the residual variation may be local. The Guyana Shield area has consistently negative residuals, showing that this area has lower tree species-richness than expected by our models. We provide extensive plot meta-data, including tree density, tree alpha-diversity and tree species-richness results and gridded maps at 0.1-degree resolution.

Published
2023
Full Text
View/download PDF

26. Overlap-aware low-latency online speaker diarization based on end-to-end local segmentation

Author

Coria, Juan M., Bredin, Hervé, Ghannay, Sahar, and Rosset, Sophie

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Sound
Abstract

We propose to address online speaker diarization as a combination of incremental clustering and local diarization applied to a rolling buffer updated every 500ms. Every single step of the proposed pipeline is designed to take full advantage of the strong ability of a recently proposed end-to-end overlap-aware segmentation to detect and separate overlapping speakers. In particular, we propose a modified version of the statistics pooling layer (initially introduced in the x-vector architecture) to give less weight to frames where the segmentation model predicts simultaneous speakers. Furthermore, we derive cannot-link constraints from the initial segmentation step to prevent two local speakers from being wrongfully merged during the incremental clustering step. Finally, we show how the latency of the proposed approach can be adjusted between 500ms and 5s to match the requirements of a particular use case, and we provide a systematic analysis of the influence of latency on the overall performance (on AMI, DIHARD and VoxConverse)., Comment: To appear in ASRU 2021. Code available at https://github.com/juanmc2005/StreamingSpeakerDiarization/

Published
2021

28. Mapping density, diversity and species-richness of the Amazon tree flora

Author

ter Steege, Hans, Pitman, Nigel C. A., do Amaral, Iêda Leão, de Souza Coelho, Luiz, de Almeida Matos, Francisca Dionízia, de Andrade Lima Filho, Diógenes, Salomão, Rafael P., Wittmann, Florian, Castilho, Carolina V., Guevara, Juan Ernesto, Veiga Carim, Marcelo de Jesus, Phillips, Oliver L., Magnusson, William E., Sabatier, Daniel, Revilla, Juan David Cardenas, Molino, Jean-François, Irume, Mariana Victória, Martins, Maria Pires, da Silva Guimarães, José Renan, Ramos, José Ferreira, Bánki, Olaf S., Piedade, Maria Teresa Fernandez, Cárdenas López, Dairon, Rodrigues, Domingos de Jesus, Demarchi, Layon O., Schöngart, Jochen, Almeida, Everton José, Barbosa, Luciane Ferreira, Cavalheiro, Larissa, dos Santos, Márcia Cléia Vilela, Luize, Bruno Garcia, de Leão Novo, Evlyn Márcia Moraes, Vargas, Percy Núñez, Silva, Thiago Sanna Freire, Venticinque, Eduardo Martins, Manzatto, Angelo Gilberto, Reis, Neidiane Farias Costa, Terborgh, John, Casula, Katia Regina, Honorio Coronado, Euridice N., Monteagudo Mendoza, Abel, Montero, Juan Carlos, Costa, Flávia R. C., Feldpausch, Ted R., Quaresma, Adriano Costa, Castaño Arboleda, Nicolás, Zartman, Charles Eugene, Killeen, Timothy J., Marimon, Beatriz S., Marimon-Junior, Ben Hur, Vasquez, Rodolfo, Mostacedo, Bonifacio, Assis, Rafael L., Baraloto, Chris, do Amaral, Dário Dantas, Engel, Julien, Petronelli, Pascal, Castellanos, Hernán, de Medeiros, Marcelo Brilhante, Simon, Marcelo Fragomeni, Andrade, Ana, Camargo, José Luís, Laurance, William F., Laurance, Susan G. W., Maniguaje Rincón, Lorena, Schietti, Juliana, Sousa, Thaiane R., de Sousa Farias, Emanuelle, Lopes, Maria Aparecida, Magalhães, José Leonardo Lima, Nascimento, Henrique Eduardo Mendonça, de Queiroz, Helder Lima, Aymard C., Gerardo A., Brienen, Roel, Stevenson, Pablo R., Araujo-Murakami, Alejandro, Baker, Tim R., Cintra, Bruno Barçante Ladvocat, Feitosa, Yuri Oliveira, Mogollón, Hugo F., Duivenvoorden, Joost F., Peres, Carlos A., Silman, Miles R., Ferreira, Leandro Valle, Lozada, José Rafael, Comiskey, James A., Draper, Freddie C., de Toledo, José Julio, Damasco, Gabriel, García-Villacorta, Roosevelt, Lopes, Aline, Vicentini, Alberto, Cornejo Valverde, Fernando, Alonso, Alfonso, Arroyo, Luzmila, Dallmeier, Francisco, Gomes, Vitor H. F., Jimenez, Eliana M., Neill, David, Peñuela Mora, Maria Cristina, Noronha, Janaína Costa, de Aguiar, Daniel P. P., Barbosa, Flávia Rodrigues, Bredin, Yennie K., de Sá Carpanedo, Rainiellen, Carvalho, Fernanda Antunes, de Souza, Fernanda Coelho, Feeley, Kenneth J., Gribel, Rogerio, Haugaasen, Torbjørn, Hawes, Joseph E., Pansonato, Marcelo Petratti, Ríos Paredes, Marcos, Barlow, Jos, Berenguer, Erika, da Silva, Izaias Brasil, Ferreira, Maria Julia, Ferreira, Joice, Fine, Paul V. A., Guedes, Marcelino Carneiro, Levis, Carolina, Licona, Juan Carlos, Villa Zegarra, Boris Eduardo, Vos, Vincent Antoine, Cerón, Carlos, Durgante, Flávia Machado, Fonty, Émile, Henkel, Terry W., Householder, John Ethan, Huamantupa-Chuquimaco, Isau, Pos, Edwin, Silveira, Marcos, Stropp, Juliana, Thomas, Raquel, Daly, Doug, Dexter, Kyle G., Milliken, William, Molina, Guido Pardo, Pennington, Toby, Vieira, Ima Célia Guimarães, Weiss Albuquerque, Bianca, Campelo, Wegliane, Fuentes, Alfredo, Klitgaard, Bente, Pena, José Luis Marcelo, Tello, J. Sebastián, Vriesendorp, Corine, Chave, Jerome, Di Fiore, Anthony, Hilário, Renato Richard, de Oliveira Pereira, Luciana, Phillips, Juan Fernando, Rivas-Torres, Gonzalo, van Andel, Tinde R., von Hildebrand, Patricio, Balee, William, Barbosa, Edelcilio Marques, de Matos Bonates, Luiz Carlos, Dávila Doza, Hilda Paulette, Zárate Gómez, Ricardo, Gonzales, Therany, Gallardo Gonzales, George Pepe, Hoffman, Bruce, Junqueira, André Braga, Malhi, Yadvinder, de Andrade Miranda, Ires Paula, Pinto, Linder Felipe Mozombite, Prieto, Adriana, Rudas, Agustín, Ruschel, Ademir R., Silva, Natalino, Vela, César I. A., Zent, Egleé L., Zent, Stanford, Cano, Angela, Carrero Márquez, Yrma Andreina, Correa, Diego F., Costa, Janaina Barbosa Pedrosa, Flores, Bernardo Monteiro, Galbraith, David, Holmgren, Milena, Kalamandeen, Michelle, Lobo, Guilherme, Torres Montenegro, Luis, Nascimento, Marcelo Trindade, Oliveira, Alexandre A., Pombo, Maihyra Marina, Ramirez-Angulo, Hirma, Rocha, Maira, Scudeller, Veridiana Vizoni, Sierra, Rodrigo, Tirado, Milton, Umaña, Maria Natalia, van der Heijden, Geertje, Vilanova Torre, Emilio, Reategui, Manuel Augusto Ahuite, Baider, Cláudia, Balslev, Henrik, Cárdenas, Sasha, Casas, Luisa Fernanda, Endara, María José, Farfan-Rios, William, Ferreira, Cid, Linares-Palomino, Reynaldo, Mendoza, Casimiro, Mesones, Italo, Parada, Germaine Alexander, Torres-Lezama, Armando, Urrego Giraldo, Ligia Estela, Villarroel, Daniel, Zagt, Roderick, Alexiades, Miguel N., de Oliveira, Edmar Almeida, Garcia-Cabrera, Karina, Hernandez, Lionel, Cuenca, Walter Palacios, Pansini, Susamar, Pauletto, Daniela, Ramirez Arevalo, Freddy, Sampaio, Adeilza Felipe, Valderrama Sandoval, Elvis H., Gamarra, Luis Valenzuela, Levesley, Aurora, Pickavance, Georgia, and Melgaço, Karina

Published
2023
Full Text
View/download PDF

30. End-to-end speaker segmentation for overlap-aware resegmentation

Author

Bredin, Hervé and Laurent, Antoine

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Sound
Abstract

Speaker segmentation consists in partitioning a conversation between one or more speakers into speaker turns. Usually addressed as the late combination of three sub-tasks (voice activity detection, speaker change detection, and overlapped speech detection), we propose to train an end-to-end segmentation model that does it directly. Inspired by the original end-to-end neural speaker diarization approach (EEND), the task is modeled as a multi-label classification problem using permutation-invariant training. The main difference is that our model operates on short audio chunks (5 seconds) but at a much higher temporal resolution (every 16ms). Experiments on multiple speaker diarization datasets conclude that our model can be used with great success on both voice activity detection and overlapped speech detection. Our proposed model can also be used as a post-processing step, to detect and correctly assign overlapped speech regions. Relative diarization error rate improvement over the best considered baseline (VBx) reaches 17% on AMI, 13% on DIHARD 3, and 13% on VoxConverse., Comment: Camera-ready version for Interspeech 2021 with significantly better voice activity detection, overlapped speech detection, and speaker diarization results. The code used for results reported in v1 contained a small bug that has now been fixed

Published
2021

37. Immediate crossover fatigue after unilateral submaximal eccentric contractions of the knee flexors involves peripheral alterations and increased global perceived fatigue.

Author

Jennifer Gioda, Flavio Da Silva, Florian Monjo, Baptiste Corcelle, Jonathan Bredin, Enzo Piponnier, and Serge S Colson

Subjects
Medicine, Science
Abstract

After a unilateral muscle exercise, the performance of the non-exercised contralateral limb muscle can be also impaired. This crossover fatigue phenomenon is still debated in the literature and very few studies have investigated the influence of eccentric contractions. This study was designed to assess neuromuscular adaptations involved in the crossover fatigue of the non-exercised contralateral knee flexor muscles. Seventeen healthy young men performed a unilateral submaximal eccentric exercise of the right knee flexors until a 20% reduction in maximal voluntary isometric contraction torque was attained in the exercised limb. Before (PRE), immediately after exercise cessation (POST) and 24 hours later (POST24), neuromuscular function and perceived muscle soreness were measured in both the exercised limb and non-exercised limb. In addition, global perceived fatigue was assessed at each measurement time. At POST, significant reductions in maximal voluntary isometric contraction were observed in the exercised limb (-28.1%, p < 0.001) and in the non-exercised limb (-8.5%, p < 0.05), evidencing crossover fatigue. At POST, voluntary activation decreased in the exercised limb only (-6.0%, p < 0.001), while electrically evoked potentiated doublet torque was impaired in both the exercised limb and the non-exercised limb (-11.6%, p = 0.001). In addition, global perceived fatigue significantly increased at POST (p < 0.001). At POST24, all measured variables returned to PRE values, except for perceived muscle soreness scores exhibiting greater values than PRE (p < 0.05). A possible cumulative interaction between peripheral alterations and global perceived fatigue may account for the immediate crossover fatigue observed in the non-exercised limb.

Published
2024
Full Text
View/download PDF

38. Reproducibility of `COST Reference Microplasma Jets'

Author

Riedel, F, Golda, J, Held, J, Davies, H, van der Woude, M W, Bredin, J, Niemi, K, Gans, T, der Gathen, V Schulz-von, and O'Connell, D

Subjects
Physics - Plasma Physics
Abstract

Atmospheric pressure plasmas have been ground-breaking for plasma science and technologies, due to their significant application potential in many fields, including medicinal, biological, and environmental applications. This is predominantly due to their efficient production and delivery of chemically reactive species under ambient conditions. One of the challenges in progressing the field is comparing plasma sources and results across the community and the literature. To address this a reference plasma source was established during the `Biomedical Applications of Atmospheric Pressure Plasmas' EU COST Action MP1101. It is crucial that reference sources are reproducible. Here, we present the reproducibility and variance across multiple sources through examining various characteristics, including: absolute atomic oxygen densities, absolute ozone densities, electrical characteristics, optical emission spectroscopy, temperature measurements, and bactericidal activity. The measurements demonstrate that the tested COST jets are reproducible, within the standard deviation, for each measurement.

Published
2020
Full Text
View/download PDF

39. An open-source voice type classifier for child-centered daylong recordings

Author

Lavechin, Marvin, Bousbib, Ruben, Bredin, Hervé, Dupoux, Emmanuel, and Cristia, Alejandrina

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, I.2.7
Abstract

Spontaneous conversations in real-world settings such as those found in child-centered recordings have been shown to be amongst the most challenging audio files to process. Nevertheless, building speech processing models handling such a wide variety of conditions would be particularly useful for language acquisition studies in which researchers are interested in the quantity and quality of the speech that children hear and produce, as well as for early diagnosis and measuring effects of remediation. In this paper, we present our approach to designing an open-source neural network to classify audio segments into vocalizations produced by the child wearing the recording device, vocalizations produced by other children, adult male speech, and adult female speech. To this end, we gathered diverse child-centered corpora which sums up to a total of 260 hours of recordings and covers 10 languages. Our model can be used as input for downstream tasks such as estimating the number of words produced by adult speakers, or the number of linguistic units produced by children. Our architecture combines SincNet filters with a stack of recurrent layers and outperforms by a large margin the state-of-the-art system, the Language ENvironment Analysis (LENA) that has been used in numerous child language studies., Comment: accepted to Interspeech 2020

Published
2020

40. The formation of atomic oxygen and hydrogen in atmospheric pressure plasmas containing humidity: picosecond two-photon absorption laser induced fluorescence and numerical simulations

Author

Schröter, Sandra, Bredin, Jérôme, Gibson, Andrew R., West, Andrew, Dedrick, James P., Wagenaars, Erik, Niemi, Kari, Gans, Timo, and O'Connell, Deborah

Subjects
Physics - Plasma Physics
Abstract

Atmospheric pressure plasmas are effective sources for reactive species, making them applicable for industrial and biomedical applications. We quantify ground-state densities of key species, atomic oxygen (O) and hydrogen (H), produced from admixtures of water vapour (up to 0.5%) to the helium feed gas in a radio-frequency-driven plasma at atmospheric pressure. Absolute density measurements, using two-photon absorption laser induced fluorescence, require accurate effective excited state lifetimes. For atmospheric pressure plasmas, picosecond resolution is needed due to the rapid collisional de-excitation of excited states. These absolute O and H density measurements, at the nozzle of the plasma jet, are used to benchmark a plug-flow, 0D chemical kinetics model, for varying humidity content, to further investigate the main formation pathways of O and H. It is found that impurities can play a crucial role for the production of O at small molecular admixtures. Hence, for controllable reactive species production, purposely admixed molecules to the feed gas is recommended, as opposed to relying on ambient molecules. The controlled humidity content was also identified as an effective tailoring mechanism for the O/H ratio., Comment: 35 pages, 14 figures

Published
2020
Full Text
View/download PDF

41. A Comparison of Metric Learning Loss Functions for End-To-End Speaker Verification

Author

Coria, Juan M., Bredin, Hervé, Ghannay, Sahar, and Rosset, Sophie

Subjects
Computer Science - Machine Learning, Computer Science - Sound, Electrical Engineering and Systems Science - Audio and Speech Processing, Statistics - Machine Learning
Abstract

Despite the growing popularity of metric learning approaches, very little work has attempted to perform a fair comparison of these techniques for speaker verification. We try to fill this gap and compare several metric learning loss functions in a systematic manner on the VoxCeleb dataset. The first family of loss functions is derived from the cross entropy loss (usually used for supervised classification) and includes the congenerous cosine loss, the additive angular margin loss, and the center loss. The second family of loss functions focuses on the similarity between training samples and includes the contrastive loss and the triplet loss. We show that the additive angular margin loss function outperforms all other loss functions in the study, while learning more robust representations. Based on a combination of SincNet trainable features and the x-vector architecture, the network used in this paper brings us a step closer to a really-end-to-end speaker verification system, when combined with the additive angular margin loss, while still being competitive with the x-vector baseline. In the spirit of reproducible research, we also release open source Python code for reproducing our results, and share pretrained PyTorch models on torch.hub that can be used either directly or after fine-tuning.

Published
2020

42. Speaker detection in the wild: Lessons learned from JSALT 2019

Author

Garcia, Paola, Villalba, Jesus, Bredin, Herve, Du, Jun, Castan, Diego, Cristia, Alejandrina, Bullock, Latane, Guo, Ling, Okabe, Koji, Nidadavolu, Phani Sankar, Kataria, Saurabh, Chen, Sizhu, Galmant, Leo, Lavechin, Marvin, Sun, Lei, Gill, Marie-Philippe, Ben-Yair, Bar, Abdoli, Sajjad, Wang, Xin, Bouaziz, Wassim, Titeux, Hadrien, Dupoux, Emmanuel, Lee, Kong Aik, and Dehak, Najim

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Sound
Abstract

This paper presents the problems and solutions addressed at the JSALT workshop when using a single microphone for speaker detection in adverse scenarios. The main focus was to tackle a wide range of conditions that go from meetings to wild speech. We describe the research threads we explored and a set of modules that was successful for these scenarios. The ultimate goal was to explore speaker detection; but our first finding was that an effective diarization improves detection, and not having a diarization stage impoverishes the performance. All the different configurations of our research agree on this fact and follow a main backbone that includes diarization as a previous stage. With this backbone, we analyzed the following problems: voice activity detection, how to deal with noisy signals, domain mismatch, how to improve the clustering; and the overall impact of previous stages in the final speaker detection. In this paper, we show partial results for speaker diarizarion to have a better understanding of the problem and we present the final results for speaker detection., Comment: Submitted to ICASSP 2020

Published
2019

43. The Speed Submission to DIHARD II: Contributions & Lessons Learned

Author

Sahidullah, Md, Patino, Jose, Cornell, Samuele, Yin, Ruiqing, Sivasankaran, Sunit, Bredin, Hervé, Korshunov, Pavel, Brutti, Alessio, Serizel, Romain, Vincent, Emmanuel, Evans, Nicholas, Marcel, Sébastien, Squartini, Stefano, and Barras, Claude

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Machine Learning, Computer Science - Sound
Abstract

This paper describes the speaker diarization systems developed for the Second DIHARD Speech Diarization Challenge (DIHARD II) by the Speed team. Besides describing the system, which considerably outperformed the challenge baselines, we also focus on the lessons learned from numerous approaches that we tried for single and multi-channel systems. We present several components of our diarization system, including categorization of domains, speech enhancement, speech activity detection, speaker embeddings, clustering methods, resegmentation, and system fusion. We analyze and discuss the effect of each such component on the overall diarization performance within the realistic settings of the challenge.

Published
2019

44. pyannote.audio: neural building blocks for speaker diarization

Author

Bredin, Hervé, Yin, Ruiqing, Coria, Juan Manuel, Gelly, Gregory, Korshunov, Pavel, Lavechin, Marvin, Fustes, Diego, Titeux, Hadrien, Bouaziz, Wassim, and Gill, Marie-Philippe

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Sound
Abstract

We introduce pyannote.audio, an open-source toolkit written in Python for speaker diarization. Based on PyTorch machine learning framework, it provides a set of trainable end-to-end neural building blocks that can be combined and jointly optimized to build speaker diarization pipelines. pyannote.audio also comes with pre-trained models covering a wide range of domains for voice activity detection, speaker change detection, overlapped speech detection, and speaker embedding -- reaching state-of-the-art performance for most of them., Comment: Submitted to ICASSP 2020

Published
2019

45. Rethinking Measurement Standards of Autism Symptomology in Adolescents with Fragile X Syndrome

Author

Fielding-Gebhardt, Heather, Bredin-Oja, Shelley L., Warren, Steven F., and Brady, Nancy C.

Abstract

Accurate representation of autism spectrum disorder (ASD) in fragile X syndrome (FXS) is necessary for the field. We examined classifications of ASD using three approaches--Autism Diagnostic Observation Schedule (ADOS-2; Lord et al. 2012), Childhood Autism Rating Scale (CARS2-ST; Schopler et al. 2010), and Vineland Adaptive Behavior Scales (VABS-II; Sparrow et al. 2005)--in 45 adolescents with FXS. Maladaptive items from the VABS-II plus a maternal interview were matched with Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5; American Psychiatric Association 2013) criteria for ASD. Results indicated discrepant classifications. The ADOS-2 yielded the highest rate of comorbid autism (71%); CARS2-ST and VABS-II/DSM-5 yielded lower rates (38% and 42%, respectively). A singular measure of autism symptomology is insufficient to characterize autism in FXS.

Published
2021
Full Text
View/download PDF

46. Overlap-aware diarization: resegmentation using neural end-to-end overlapped speech detection

Author

Bullock, Latané, Bredin, Hervé, and Garcia-Perera, Leibny Paola

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Sound
Abstract

We address the problem of effectively handling overlapping speech in a diarization system. First, we detail a neural Long Short-Term Memory-based architecture for overlap detection. Secondly, detected overlap regions are exploited in conjunction with a frame-level speaker posterior matrix to make two-speaker assignments for overlapped frames in the resegmentation step. The overlap detection module achieves state-of-the-art performance on the AMI, DIHARD, and ETAPE corpora. We apply overlap-aware resegmentation on AMI, resulting in a 20% relative DER reduction over the baseline system. While this approach is by no means an end-all solution to overlap-aware diarization, it reveals promising directions for handling overlap., Comment: Submitted to ICASSP 2020

Published
2019

47. End-to-end Domain-Adversarial Voice Activity Detection

Author

Lavechin, Marvin, Gill, Marie-Philippe, Bousbib, Ruben, Bredin, Hervé, and Garcia-Perera, Leibny Paola

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, I.2.7
Abstract

Voice activity detection is the task of detecting speech regions in a given audio stream or recording. First, we design a neural network combining trainable filters and recurrent layers to tackle voice activity detection directly from the waveform. Experiments on the challenging DIHARD dataset show that the proposed end-to-end model reaches state-of-the-art performance and outperforms a variant where trainable filters are replaced by standard cepstral coefficients. Our second contribution aims at making the proposed voice activity detection model robust to domain mismatch. To that end, a domain classification branch is added to the network and trained in an adversarial manner. The same DIHARD dataset, drawn from 11 different domains is used for evaluation under two scenarios. In the in-domain scenario where the training and test sets cover the exact same domains, we show that the domain-adversarial approach does not degrade performance of the proposed end-to-end model. In the out-domain scenario where the test domain is different from training domains, it brings a relative improvement of more than 10%. Finally, our last contribution is the provision of a fully reproducible open-source pipeline than can be easily adapted to other datasets., Comment: submitted to Interspeech 2020

Published
2019

48. LSTM based Similarity Measurement with Spectral Clustering for Speaker Diarization

Author

Lin, Qingjian, Yin, Ruiqing, Li, Ming, Bredin, Hervé, and Barras, Claude

Subjects
Electrical Engineering and Systems Science - Audio and Speech Processing, Computer Science - Machine Learning, Computer Science - Sound, Statistics - Machine Learning
Abstract

More and more neural network approaches have achieved considerable improvement upon submodules of speaker diarization system, including speaker change detection and segment-wise speaker embedding extraction. Still, in the clustering stage, traditional algorithms like probabilistic linear discriminant analysis (PLDA) are widely used for scoring the similarity between two speech segments. In this paper, we propose a supervised method to measure the similarity matrix between all segments of an audio recording with sequential bidirectional long short-term memory networks (Bi-LSTM). Spectral clustering is applied on top of the similarity matrix to further improve the performance. Experimental results show that our system significantly outperforms the state-of-the-art methods and achieves a diarization error rate of 6.63% on the NIST SRE 2000 CALLHOME database., Comment: Accepted for INTERSPEECH 2019

Published
2019
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results