ROMPAR: Morphological Completion and Demographic Unlearning for Romanian-Accented Speech Recognition
It addresses demographic bias and technical artifacts in parliamentary ASR for Romanian and Moldavian, a low-resource dialectal setting.
The paper introduces the ROMPAR dataset (17.80 hours of Romanian/Moldavian parliamentary speech) and proposes a multi-task adversarial training framework with exponential decay and LLM-guided decoding for morphological completion, achieving 96.6% F1-score in reconstruction and significant WER reduction.
Automated transcription of parliamentary proceedings faces significant hurdles due to demographic bias, dialectal variation, and technical artifacts such as utterance truncation during segmentation. This paper introduces the ROManian PARliamentary Speech Corpus (ROMPAR) dataset, a 17.80-hour corpus of Romanian and Moldavian parliamentary speech, featuring double-annotated ground truth and explicit labels for reconstructed word fragments. To build a robust ASR system, we propose a multi-task adversarial training framework that enforces demographic invariance across age, gender, and dialect. We address the inherent instability of adversarial objectives in generative architectures by introducing an exponential decay mechanism for the adversarial coefficients. Furthermore, we implement an LLM-guided decoding strategy with position-dependent weighting to facilitate morphological completion of truncated terminal words. Our results demonstrate that the proposed framework significantly reduces WER and achieves an F1-score of 96.6% in morphological reconstruction.