Diarization-Guided Qwen-ASR Adaptation for Multilingual Two-Speaker Conversational Speech
Quick Answer
The proposed system for the MLC-SLM 2026 Challenge integrates a modular speaker diarization front end with Qwen3-ASR-1.7B, achieving a tcpMER of 17.97 on the evaluation set.
Quick Take
The proposed system for the MLC-SLM 2026 Challenge integrates a modular speaker diarization front end with Qwen3-ASR-1.7B, achieving a tcpMER of 17.97 on the evaluation set. The approach includes supervised fine-tuning, LoRA adaptation with synthetic speech, and GRPO reinforcement learning, resulting in a 6.83 absolute point reduction in error rate. This advancement significantly enhances multilingual two-speaker conversational speech recognition.
Key Points
- System combines speaker diarization with Qwen3-ASR-1.7B for improved recognition.
- Achieved a tcpMER of 17.97 on the final evaluation set.
- Supervised fine-tuning provided the largest performance gain.
- Synthetic speech LoRA adaptation and GRPO reinforcement learning enhanced robustness.
- Error rate reduced by 6.83 absolute points from baseline performance.
Paper Resources
📖 Reader Mode
~2 min readAbstract:This paper describes our self-designed system for Task 1 of the MLC-SLM 2026 Challenge for multilingual two-speaker conversational speech. The system combines a modular speaker diarization front end with a challenge-adapted Qwen3-ASR-1.7B recognizer. The diarization front end performs voice activity detection, subsegment generation, CAMPPlus speaker embedding extraction, two-speaker spectral clustering, and RTTM-based audio segmentation. The resulting speaker-attributed segments are grouped by language or region and decoded by the adapted ASR model. For ASR adaptation, we first perform supervised full fine-tuning on the official training data, then apply LoRA fine-tuning with synthetic speech generated by a three-pipeline TTS-based synthetic speech augmentation framework, and finally refine the model using GRPO reinforcement learning with rewards based on WER/CER and penalties for hallucination, repetition, and length deviation. On the official development set, the full system achieves an average tcpMER of 23.70, reducing the error rate by 6.83 absolute points relative to the released Qwen-ASR-1.7B performance. On the final evaluation set, the system achieves an average tcpMER of 17.97. Ablation results show that supervised fine-tuning provides the largest gain, while synthetic-speech LoRA adaptation and reinforcement learning further improve robustness.
| Comments: | 4 main pages plus 1 page of reference |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.08208 [cs.CL] |
| (or arXiv:2607.08208v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.08208 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Hao Wu [view email]
[v1]
Thu, 9 Jul 2026 08:07:34 UTC (32 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Quantifying Prior Dominance in Systems
The study introduces the Normalized Context Utilization (NCU) metric to evaluate Retrieval-Augmented Generation (RAG) systems, revealing that Small Language Models (SLMs) outperform larger models in factual extraction. The findings indicate that traditional scaling laws yield diminishing returns, with a commercial API frequently failing against adversarial evidence due to systemic confidence collapse.