Federated learning in audio processing presents distinct challenges, such as data heterogeneity across devices leading to inconsistencies in audio quality and content, which can affect model reliability. Issues like temporal misalignment due to variations in sampling rates and delays in audio recordings complicate training. These are exacerbated by diverse computational resources and storage capacities on different devices, impacting model efficacy. Additionally, privacy concerns arise with managing large-scale audio data, necessitating robust encryption for data protection. The asynchronous nature of device connectivity, potential clock drift, and network latency can further delay model updates, affecting performance. Addressing these challenges requires the development of adaptable algorithms, enhanced security protocols, and efficient strategies for connectivity and data synchronization.

In this workshop at IEEE ICASSP 2025, we aim to facilitate the sharing of cutting-edge research, foster collaborations among scholars, and drive forward the practical applications of federated learning in audio processing.

Call for Paper

The workshop aims to provide a platform for exchanging ideas on the future of federated learning surrounding audio applications. The main topics of interest include, but are not limited to:

Topics of interest include, but are not limited to:

Applications of FL in the Audio Domain
Automatic Speech Recognition
Audio, Sound, and Music Processing
Personalized Experiences
Multi-device, Multi-Modal, and Self-supervised FL
Enhancing Healthcare
FL and LLMs for Advanced Audio Understanding
Prompt Tuning in FL Settings
FL Frameworks for Foundation Models
Exploiting LLM Embeddings for Audio Applications
Efficient FL for Audio Tasks
Addressing User and Device Heterogeneity
Resource-Constrained FL
Adaptive Aggregation Strategies
Energy Efficiency
Robustness, Bias, and Interpretability in FL for Audio
Security and Privacy for Audio Tasks
Bias and Interpretability
Handling Cross-Domain Data
Federated Unlearning

Submission Guidelines:

FLute 2025 will offer two primary submission tracks, both of which will be subject to a peer-review process. Works accepted in these tracks will be selected based on their technical quality and overall contribution to the event. The key distinctions between the tracks are outlined below:

(A) Full Paper Track submissions should present well-developed, cohesive research with significant technical depth and clear, impactful relevance to federated learning for audio understanding. Accepted papers in this track will be published in the IEEE Xplore Digital Library. Full papers should abide by the ICASSP-2025 paper format.

(B) Extended Abstract Track encourages submissions that promote insight and engagement through innovative ideas, discussions, resource sharing, and fostering collaborations. It welcomes non-traditional research such as novel datasets, negative results, preliminary findings, reproducibility studies, and opinion pieces. Submissions can be extended abstracts up to 2 pages, references included. While these abstracts will not be published in the IEEE Xplore Digital Library, accepted authors will have the option (but are not required) to submit their work to arxiv.org and have it linked on the workshop’s website.