Publications/Efficient Machine Learning: Data Selection, Pruning, Active & Federated Learning
Journal · 2025
A Survey on Data Selection for Efficient Speech Processing
Abdul Hameed Azeemi, Ihsan Ayyub Qazi, Agha Ali Raza
IEEE Access, vol. 13, 2025 · IF 4.2
Summary
Modern speech models are powerful but expensive to train. One remedy is to train on less, but better chosen, data. This survey organizes the growing body of data selection methods for speech into a single taxonomy along three dimensions: what is selected (whole samples or segments), when (once, during training, or through active learning), and by what criterion (uncertainty, diversity, or both). It reviews how these strategies perform across speech recognition, text-to-speech, audio anti-spoofing, speaker recognition, and emotion recognition.
Cite this paper
Abdul Hameed Azeemi, Ihsan Ayyub Qazi, Agha Ali Raza. A Survey on Data Selection for Efficient Speech Processing. IEEE Access, vol. 13, 2025.
@article{azeemi2025survey,
author = {Abdul Hameed Azeemi and Ihsan Ayyub Qazi and Agha Ali Raza},
title = {{A Survey on Data Selection for Efficient Speech Processing}},
journal = {IEEE Access},
volume = {13},
year = {2025},
url = {https://doi.org/10.1109/ACCESS.2025.3582395}
}
- Language Model-Driven Data Pruning Enables Efficient Active LearningFindings of EACL 2026, Rabat, Morocco
- To Label or Not to Label: Hybrid Active Learning for Neural Machine TranslationCOLING 2025, Abu Dhabi, UAE
- Data Pruning for Efficient Model Pruning in Neural Machine TranslationFindings of EMNLP 2023, Singapore
- Representative Subset Selection for Efficient Fine-Tuning in Self-Supervised Speech RecognitionNeurIPS 2023 Workshop (ENLSP), New Orleans, USA
- Learning Fast and Slow: Towards Inclusive Federated LearningECML PKDD 2023, Turin, Italy