Technology Category
Natural Language Processing: Conversational AI/Dialog SystemsData Mining & Knowledge Management: Conversational Systems for Recommendation & RetrievalMachine Learning: Large Multimodal Models (LMMs)
Toward Conversational Hungarian Speech Recognition: Introducing the BEA-Large and BEA-Dialogue Datasets
🤖 AI Summary
To address the critical bottleneck of scarce spontaneous, conversational speech data for low-resource languages like Hungarian—hindering ASR advancement—this work constructs and publicly releases two high-quality Hungarian conversational speech datasets: BEA-Large (255 hours, containing both spontaneous and read speech) and BEA-Dialogue (85 hours, real multi-turn dialogues), the first to provide fine-grained segmentation annotations and speaker-independent splits. Leveraging a Fast Conformer architecture, fine-tuned models achieve word error rates of 14.18% on spontaneous speech and 4.8% on read speech on BEA-Large; end-to-end speaker diarization error rates range from 13.05% to 18.26%, confirming dataset difficulty and benchmark utility. These contributions fill a fundamental gap in conversational ASR resources for low-resource languages and enable reproducible research.
Technology Category
Application Category
📝 Abstract
The advancement of automatic speech recognition (ASR) has been largely enhanced by extensive datasets in high-resource languages, while languages such as Hungarian remain underrepresented due to limited spontaneous and conversational corpora. To address this gap, we introduce two new datasets -- BEA-Large and BEA-Dialogue -- constructed from the previously unprocessed portions of the Hungarian speech corpus named BEA. BEA-Large extends BEA-Base with 255 hours of spontaneous speech from 433 speakers, enriched with detailed segment-level metadata. BEA-Dialogue, comprising 85 hours of spontaneous conversations, is a Hungarian speech corpus featuring natural dialogues partitioned into speaker-independent subsets, supporting research in conversational ASR and speaker diarization. We establish reproducible baselines on these datasets using publicly available ASR models, with the fine-tuned Fast Conformer model achieving word error rates as low as 14.18% on spontaneous and 4.8% on repeated speech. Diarization experiments yield diarization error rates between 13.05% and 18.26%, providing reference points for future improvements. The results highlight the persistent difficulty of conversational ASR, particularly due to disfluencies, overlaps, and informal speech patterns. By releasing these datasets and baselines, we aim to advance Hungarian speech technology and offer a methodological framework for developing spontaneous and conversational benchmarks in other languages.
Problem
Research questions and friction points this paper is trying to address.
Hungarian lacks spontaneous conversational speech datasets for ASR development
Existing Hungarian speech corpora are insufficient for conversational ASR research
Limited resources hinder speaker diarization and informal speech recognition in Hungarian
Innovation
Methods, ideas, or system contributions that make the work stand out.
Introducing BEA-Large and BEA-Dialogue Hungarian speech datasets
Fine-tuning Fast Conformer model for improved recognition accuracy
Providing reproducible baselines for conversational ASR and diarization
🔎 Similar Papers
No similar papers found.
