Research map / Video / Video-language understanding
Reasoning LLM Improves Speaker Recognition in Long-form TV Dramas
Yuxuan Li, Lingxi Xie, Xinyue Huo, Jihao Qiu, Jiacheng Shao, Pengfei Chen, Jiannan Ge, Kaiwen Duan, Qi Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add RL (GRPO) after SFT | Qwen3-8B DramaSR-LRM · Speaker recognition test set | accuracy (%): 75.22% -> 86.93% | Improved |
| add SFT and RL on top of direct Qwen3-8B use | Qwen3-8B · Speaker recognition test set | accuracy (%): 27.40% -> 86.93% | Improved |
| add confidence sampling (invoke LRM only when margin < rho) | DramaSR-LRM · Speaker recognition test set | accuracy (%): 86.93% -> 87.79% | Improved |
| add confidence sampling to SFT-only model | Qwen3-8B + SFT · Speaker recognition test set | accuracy (%): 75.22% -> 82.70% | Improved |
| add the voice_sim tool (ablation) | DramaSR-LRM · Speaker recognition test set | accuracy (%): 72.61% -> 87.79% | Improved |
| add the video_cap tool (ablation) | DramaSR-LRM · Speaker recognition test set | accuracy (%): 86.76% -> 87.79% | Improved |
| add the char_relation tool (ablation) | DramaSR-LRM · Speaker recognition test set | accuracy (%): 87.55% -> 87.79% | Improved |
| add the video_cap tool (ablation) | DramaSR-LRM · Very short utterances (0-0.5s) | accuracy (%): 72.33% -> 76.65% | Improved |
| replace label propagation with DramaSR-LRM | Label propagation · Very short utterances (0-0.5s) | accuracy (%): 67.45% -> 76.65% | Improved |
| replace label propagation with DramaSR-LRM | Label propagation · English subset | accuracy (%): 82.41% -> 85.22% | Improved |
| add a second iterative refinement pass (ablation) | DramaSR-LRM · The Long Night | accuracy (%): 90.15% -> 90.40% | Improved |
| raise confidence threshold rho from 0.10 to 0.20 (ablation) | DramaSR-LRM · Stand by Me 1 | accuracy (%): 93.63 -> 91.91 | Hurt |
Closest papers by meaning
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.86)
- MMAR: A Challenging Benchmark for Deep Reasoning in Speech, Audio, Music, and Their Mix (NeurIPS 2025 · similarity 0.86)
- Mellow: a small audio language model for reasoning (NeurIPS 2025 · similarity 0.86)
- LiveCC: Learning Video LLM with Streaming Speech Transcription at Scale (CVPR 2025 · similarity 0.85)
- MMSU: A Massive Multi-task Spoken Language Understanding and Reasoning Benchmark (ICLR 2026 · similarity 0.85)
- Can Speech LLMs Think while Listening? (ICLR 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
