Research map / Architectures and efficiency / Attention and transformers
RobuTrans: A Robust Transformer-Based Text-to-Speech Model
Naihan Li, Yanqing Liu, Yu Wu, Shujie Liu, Sheng Zhao, Ming Liu
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 |
|---|---|---|---|
| replace encoder-decoder soft attention with duration-based hard attention plus PNCA, no position embedding | TransformerTTS · Bad-case set | bad cases: 237 -> 0 | Improved |
| RobuTrans vs TransformerTTS naturalness | RobuTrans · General set | MOS: 4.37 -> 4.36 | No clear effect |
| RobuTrans vs FastSpeech naturalness | RobuTrans · General set | CMOS: +0.187 | Improved |
| add pseudo non-causal attention (vs causal self-attention) (ablation) | RobuTrans · General set | CMOS: -0.291 CMOS when removed | Improved |
| add prosodic features (ablation) | RobuTrans · General set | CMOS: -0.134 CMOS when removed | Improved |
| add linguistic (phonemic+prosodic) features to input (ablation) | TransformerTTS · General set | CMOS: -0.047 | Hurt |
| use forward/GMM/monotonic/guided attention instead of hard duration attention | TransformerTTS · Bad-case set | bad cases | No clear effect |
Closest papers by meaning
- Neural Speech Synthesis with Transformer Network (AAAI 2019 · similarity 0.90)
- NaturalSpeech 2: Latent Diffusion Models are Natural and Zero-Shot Speech and Singing Synthesizers (ICLR 2024 · similarity 0.87)
- Mega-TTS 2: Boosting Prompting Mechanisms for Zero-Shot Speech Synthesis (ICLR 2024 · similarity 0.87)
- CLaM-TTS: Improving Neural Codec Language Model for Zero-Shot Text-to-Speech (ICLR 2024 · similarity 0.86)
- DELTA-TTS: Adapting Autoregressive Model into a Diffusion Language Model for Text-to-Speech (NeurIPS 2026 · similarity 0.86)
- MaskGCT: Zero-Shot Text-to-Speech with Masked Generative Codec Transformer (ICLR 2025 · similarity 0.86)
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.
