Swiss Parliaments Corpus Re-Imagined (SPC_R): Enhanced Transcription with RAG-based Correction and Predicted BLEU

2025-06-09

AI Summary

This study addresses the low transcription quality and semantic incompleteness of Swiss German parliamentary speech in low-resource settings. Methodologically, it introduces an ASR-LLM-filtering collaborative paradigm: (1) initial transcription using Whisper Large-v3; (2) a novel two-stage LLM-based correction framework—driven by GPT-4o and enhanced via protocol-aligned RAG—to rectify entity errors and verify semantic completeness; and (3) a dynamic quality filtering mechanism combining Predicted BLEU and GPT-4o scoring. The pipeline yields an 801-hour high-quality Swiss parliamentary speech–Standard German parallel dataset, of which 751 hours pass rigorous quality assurance. Sentence-level BLEU improves by 6 points over the original SPC baseline. This work is the first to incorporate structured protocol information into RAG retrieval and empirically validates the effectiveness of the ASR-LLM-filtering closed-loop for constructing professional, low-resource dialect corpora.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Speech
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

This paper presents a new long-form release of the Swiss Parliaments Corpus, converting entire multi-hour Swiss German debate sessions (each aligned with the official session protocols) into high-quality speech-text pairs. Our pipeline starts by transcribing all session audio into Standard German using Whisper Large-v3 under high-compute settings. We then apply a two-step GPT-4o correction process: first, GPT-4o ingests the raw Whisper output alongside the official protocols to refine misrecognitions, mainly named entities. Second, a separate GPT-4o pass evaluates each refined segment for semantic completeness. We filter out any segments whose Predicted BLEU score (derived from Whisper's average token log-probability) and GPT-4o evaluation score fall below a certain threshold. The final corpus contains 801 hours of audio, of which 751 hours pass our quality control. Compared to the original sentence-level SPC release, our long-form dataset achieves a 6-point BLEU improvement, demonstrating the power of combining robust ASR, LLM-based correction, and data-driven filtering for low-resource, domain-specific speech corpora.

Problem

• Convert multi-hour Swiss German debates into high-quality speech-text pairs
• Improve transcription accuracy using Whisper and GPT-4o correction
• Enhance corpus quality with data-driven filtering and BLEU scoring

Innovation

• Whisper Large-v3 for Standard German transcription
• GPT-4o two-step correction for misrecognitions
• Predicted BLEU and GPT-4o for quality filtering