Speaker: Sara Barahona Quirós.
Abstract: Cross-lingual speaker verification remains a challenging problem due to variability in language, recording conditions, and speaker characteristics. In this talk, we present our participation in the new benchmark TidyVoice Challenge: Cross-Lingual Speaker Verification, associated with Interspeech 2026, where the goal is to develop robust systems capable of verifying speaker identity across different languages. We describe our pipeline, including data exploration and model design, highlight also key design choices and strategies to obtain speaker embeddings robust to language mismatch, concluding with lessons learned and perspectives for future multilingual speaker verification.