This session aims to build a community of speech scientists and technologists advancing explainability methods, metrics, and evaluation frameworks in speech AI, not only as a tool for understanding model behaviour but with a focus on operationalizing explainable speech AI for accountability, compliance and trust in real-world deployment. This session is complementary to previous and existing special sessions by focusing on explainability as a foundation for compliance, evidence-building, and operational safety in regulation-critical applications, where Speech AI must meet emerging legal and societal expectations for transparency.

In line with the theme of Interspeech 2026, Speaking Together, this session brings together researchers and developers to discuss how explainability can make speech AI more trustworthy.

Topics

We encourage and hence hope to receive contributions on topics for the recommended themes such as (but not limited to):

• Explainability methods for regulated and high-risk speech AI: 

  • Post-hoc and intrinsic explanation techniques

  •  Interpretation of model behaviour and decision patterns

  • Analysis of representations and feature use in deployed models

  • Reliability and stability of explanations across conditions

• Evaluating explainability for deployment: 

  • Frameworks linking explanation quality with accountability and assurance

  • Validation of explainability under real-world operational constraints

  • Human-in-the-loop and stakeholder-oriented evaluation methods

  • Benchmarking explainability for reproducibility

• Cross-domain and human-centred approaches: 

  • Adapting explainability to diverse domains and user needs

  • Human-centred explanation design

  • Bridging explainability research with deployment practice

Format

The proposed session will run for 2 hours. It will open with a brief introduction (5 mins) to outline the session’s motivation and structure. The session will conclude with a 5-minute closing to summarise key insights and future directions. The presentation format depends on the number of accepted papers:

• 6–12 papers: three will be selected for oral presentation (45 mins total— 15 minutes each, including 10 minutes for the talk and 5 minutes for Q&A); the rest will be presented as posters (1 hour).

• More than 12 papers: the session will consist entirely of posters to ensure sufficient presentation and feedback time.
  
  

Important Dates

Paper Submission Portal Opens - 17 Jan, 2026 (24:00 AOE)

Paper Submission - 25 Feb, 2026 (24:00 AOE)

Paper Update - 4 March, 2026 (24:00 AOE)

Paper Notifications Sent - 5 June, 2026  (24:00 AOE)

Camera Ready Paper Due - 19 June, 2026  (24:00 AOE)

Organizers