Acoustic Domain
The acoustic domain encompasses the study and analysis of sound signals, focusing on extracting meaningful information and building robust systems for various applications. Current research emphasizes developing effective self-supervised learning models, often employing convolutional neural networks (CNNs) combined with state-space models (SSMs) or transformer architectures, to address challenges like limited labeled data and acoustic mismatch across different recording environments. These advancements are improving performance in diverse areas such as bioacoustic analysis, speech recognition, speaker diarization, and even fall detection using ambient audio, impacting fields ranging from wildlife conservation to forensic science and assistive technologies. The development of new benchmarks and datasets, along with exploration of techniques like domain adaptation and data augmentation, are crucial for advancing the field.
Papers
MUST&P-SRL: Multi-lingual and Unified Syllabification in Text and Phonetic Domains for Speech Representation Learning
Noé Tits
Spatial HuBERT: Self-supervised Spatial Speech Representation Learning for a Single Talker from Multi-channel Audio
Antoni Dimitriadis, Siqi Pan, Vidhyasaharan Sethu, Beena Ahmed