Research map / Image generation / Autoregressive and latent generation
Unified Segment-to-Segment Framework for Simultaneous Sequence Generation
Shaolei Zhang, Yang Feng
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 |
|---|---|---|---|
| Use Seg2Seg instead of T-T on latency | Streaming ASR · LibriSpeech | latency: 610 -> 324 | Improved |
| Use Seg2Seg instead of T-T on noisy subset | Streaming ASR · LibriSpeech other | WER (%): 9.50 -> 8.73 | Improved |
| Use Seg2Seg instead of T-T on clean subset | Streaming ASR · LibriSpeech clean | WER (%): 3.40 -> 3.55 | No clear effect |
| Use Seg2Seg instead of MoChA | Streaming ASR · LibriSpeech other | WER (%): 14.20 -> 8.73 | Improved |
| Add latency-reducing methods (SelfAlign) to T-T, lowering latency | T-T · LibriSpeech | WER other (%): 9.50 -> 10.70 | Hurt |
| Replace fixed 280ms segmentation with adaptive latent segments (ablation) | Seg2Seg · Buckeye segmentation | R-value: 38.4 -> 41.2 | Improved |
| Replace fixed 280ms segmentation with adaptive latent segments (ablation) | Seg2Seg · Buckeye segmentation | precision: 28.1 -> 41.1 | Improved |
| Replace fixed-length source segmentation with adaptive aggregation (ablation) | Seg2Seg · SimulST MuST-C En-De | BLEU: 1 BLEU | Improved |
| Replace wait-k/wait-k-stride-n emission with adaptive emission (ablation) | Seg2Seg · SimulST MuST-C En-De | BLEU | Improved |
| Add multi-task learning with streaming ASR and SimulMT | Seg2Seg · SimulST MuST-C En-De | BLEU | Improved |
| Increase latency weight lambda from 0.05 to 0.4 | Seg2Seg · SimulMT WMT15 De-En | BLEU: 32.20 -> 27.90 | Hurt |
| Increase latency weight lambda from 0.05 to 0.4 | Seg2Seg · SimulST MuST-C En-De | BLEU: 23.63 -> 14.47 | Hurt |
Closest papers by meaning
- Large Language Models Are Read/Write Policy-Makers for Simultaneous Generation (AAAI 2025 · similarity 0.87)
- Divergence-Guided Simultaneous Speech Translation (AAAI 2024 · similarity 0.84)
- High-Fidelity Simultaneous Speech-To-Speech Translation (ICML 2025 · similarity 0.84)
- SimulMEGA: MoE Routers are Advanced Policy Makers for Simultaneous Speech Translation (NeurIPS 2025 · similarity 0.84)
- Revisit Micro-batch Clipping: Adaptive Data Pruning via Gradient Manipulation (ICLR 2025 · similarity 0.83)
- Overcoming Non-monotonicity in Transducer-based Streaming Generation (ICML 2025 · similarity 0.83)
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.
