Differentiable learning and processing of temporal, spatial, spectral and time-frequency signal representations

• Grant number: 22/10909-5
• Support Opportunities: Regular Research Grants
• Start date: March 01, 2023
• End date: August 31, 2025
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: Miguel Arjona Ramírez
• Grantee: Miguel Arjona Ramírez
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated researchers: Celso Setsuo Kurashima ; Demostenes Zegarra Rodriguez ; Mario Minami ; Renata Lopes Rosa ; Wesley Beccaro
• Associated scholarship(s): 23/11258-0 - Speech quality assessment by means of machine learning for next generation networks subject to atmospheric phenomena, BP.TT

Abstract

This research plan addresses temporal and spatial signals such as speech, audio and images to be processed for learning features based on differentiable losses or objective functions. In the process, it is important to convert the signals from their original domains to other ones, resulting in spectral and time-frequency representations in order to disclose features that are more significant for specific tasks. They may be reached by direct transforms like the short-time Fourier transform (STFT) or by filtering, correlation, prediction or cepstral methods directly or by adaptive means like convolutional neural networks (CNNs), recurrent neural networks (RNNs) and transformer layers provided with attention mechanisms. For the latter cases, the representations are latent variables and interpretation or regularization methods may be necessary depending on the task at hand. Upon a common machine learning base, several tasks are envisaged such as speech analysis and synthesis, spatial speech and audio enhancement and separation, objective speech quality assessment methods based on perceptual measures and identification of regions of interest in images. (AU)

Scientific publications

(7)

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

ARAUJO, ARNALDO RAFAEL CAMARA; OKEY, OGOBUCHI DANIEL; SAADI, MUHAMMAD; ADASME, PABLO; ROSA, RENATA LOPES; RODRIGUEZ, DEMOSTENES ZEGARRA. Quantum-assisted federated intelligent diagnosis algorithm with variational training supported by 5G networks. SCIENTIFIC REPORTS, v. 14, n. 1, p. 12-pg., 2024-11-01. (22/10909-5)

CHIOZZOTTO, MAURO; RAMIREZ, MIGUEL ARJONA. What Is the Best Solution for Smart Buildings? A Case Study of Fog, Edge Computing and Smart IoT Devices. APPLIED SCIENCES-BASEL, v. 15, n. 7, p. 24-pg., 2025-03-31. (22/10909-5)

PRETO, MURILO DE SOUZA; FERREIRA, FERNANDO TEUBL; KURASHIMA, CELSO SETSUO. A Virtual Class Tool for Facial Expressions Telemonitoring through Embedded Systems. 25TH SYMPOSIUM ON VIRTUAL AND AUGMENTED REALITY, SVR 2023, v. N/A, p. 5-pg., 2023-01-01. (22/10909-5)

BECCARO, WESLEY; RAMIREZ, MIGUEL ARJONA; LIAW, WILLIAM; GUIMARAES, HEITOR RODRIGUES. Analysis of Oral Exams With Speaker Diarization and Speech Emotion Recognition: A Case Study. IEEE TRANSACTIONS ON EDUCATION, v. 67, n. 1, p. 13-pg., 2023-10-18. (22/10909-5)

BATISTA, ANDREZA P.; AYUB, MUHAMMAD S.; ADASME, PABLO; BEGAZO, DANTE C.; SHAD, MUHAMMAD R.; SAADI, MUHAMMAD; ROSA, RENATA L.; RODRIGUEZ, DEMOSTENES Z.. A methodology for estimating radiofrequency signal attenuation from rainfall and atmospheric gases in 5G-and-beyond networks. IET NETWORKS, v. 14, n. 1, p. 16-pg., 2025-01-01. (22/10909-5)

PEREIRA, PEDRO HENRIQUE; BECCARO, WESLEY; RAMIREZ, MIGUEL ARJONA. Advantages and Pitfalls of Dataset Condensation: An Approach to Keyword Spotting with Time-Frequency Representations. ELECTRONICS, v. 13, n. 11, p. 13-pg., 2024-06-01. (22/10909-5)

COSTA, VICTOR; BECCARO, WESLEY. Benefits of Intelligent Fuzzy Controllers in Comparison to Classical Methods for Adaptive Optics. PHOTONICS, v. 10, n. 9, p. 18-pg., 2023-09-01. (22/10909-5)