ROMPAR: Morphological Completion and Demographic Unlearning for Romanian-Accented Speech Recognition

2026-06-14 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

The ROMPAR (ROManian PARliamentary Speech Corpus) dataset, a 17.80-hour collection of Romanian and Moldavian parliamentary speech, is introduced to address challenges in automated transcription, including demographic bias, dialectal variation, and utterance truncation. This corpus features double-annotated ground truth and explicit labels for reconstructed word fragments. To build a robust Automatic Speech Recognition (ASR) system, a multi-task adversarial training framework is proposed, designed to enforce demographic invariance across age, gender, and dialect. The framework tackles adversarial objective instability in generative architectures by incorporating an exponential decay mechanism for adversarial coefficients. Additionally, an LLM-guided decoding strategy with position-dependent weighting is implemented to facilitate morphological completion of truncated terminal words. The system significantly reduces Word Error Rate (WER) and achieves an F1-score of 96.6% in morphological reconstruction.

Key takeaway

For Machine Learning Engineers developing ASR systems for diverse or accented speech, this research offers a robust approach to mitigate demographic bias and improve transcription accuracy. You should consider integrating multi-task adversarial training with an exponential decay mechanism to stabilize generative architectures. Furthermore, implementing an LLM-guided decoding strategy can significantly enhance morphological completion for truncated words, directly improving your system's overall Word Error Rate.

Key insights

A multi-task adversarial framework with LLM-guided decoding improves ASR for accented speech by reducing demographic bias and completing morphology.

Principles

• Enforce demographic invariance in ASR.
• Address adversarial instability with decay.
• Use LLMs for morphological completion.

Method

The proposed method uses multi-task adversarial training with exponential decay for coefficients and an LLM-guided decoding strategy with position-dependent weighting to complete truncated words.

In practice

• Apply adversarial training for bias reduction.
• Implement LLM-guided decoding for ASR.
• Create double-annotated speech corpora.

Topics

• Automatic Speech Recognition
• Demographic Bias Mitigation
• Adversarial Training
• Large Language Models
• Morphological Completion
• Speech Corpora

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Evaluation of Automatic Speech Recognition Using Generative Large Language Models — Generative LLMs significantly improve ASR evaluation by aligning better with human perception than traditional WER.
• Beyond Cascaded Pipelines: Building a Native Spoken Language Model Prototype — Native Spoken Language Models (SLMs) overcome cascaded pipeline limitations by directly processing audio, preserving nuance, reducing errors, and lowering latency.
• PolySpeech-100: A Large-Scale Benchmark for Speech Understanding Across 100+ Languages and Dialects — PolySpeech-100 reveals E2E Speech-LLMs excel on dialects, but open-source models struggle with low-resource languages, and CoT prompting can degrade performance.
• Scaling Human and G2P Supervision for Robust Phonetic Transcription — G2P supervision for phonetic transcription has diminishing returns beyond 20-30 hours of human annotation.
• Granite 4.0 1B Speech: Compact, Multilingual, and Built for the Edge — Granite 4.0 1B Speech offers high accuracy and multilingual support in a compact, efficient model for edge devices.
• Contrastive Training with LLM-generated Near-Misses for Robust Code-Switching Speech Recognition — Contrastive training with LLM-generated near-misses robustly improves code-switching ASR at critical language-alternation points.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.