Research map / Image generation / Autoregressive and latent generation
Flowtron: an Autoregressive Flow-based Generative Network for Text-to-Speech Synthesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Increase prior variance from 0.0 to 0.5 or 1.0 (ablation) | Flowtron · Sally speech synthesis | speech variation | Helped |
| Set prior variance to 0.0 (ablation) | Flowtron · Sally speech synthesis | duration variation | Hurt |
| Increase prior variance (ablation) | Flowtron · Sally speech synthesis | $F_0$ contour variation | Helped |
| Vary Prenet dropout probability around 0.5 (ablation) | Tacotron 2 · Sally speech synthesis | $F_0$ contour variation | No clear effect |
| Sample the posterior from reference samples with high $F_0$ variance | Flowtron · Make monotonic Sally more expressive using Helen references | expressiveness | Helped |
| Condition on a reference sample or averaged style token | Tacotron 2 GST · Make monotonic Sally more expressive using Helen references | expressiveness | No clear effect |
| Transfer posterior style and alignment from a female speaker to a male speaker | Flowtron · LibriTTS seen speakers | similarity to source speaker's acoustic characteristics | Helped |
| Condition on unseen somber-style reference audio | Flowtron · Sally, Born of Darkness evaluation samples | style transfer | Helped |
| Condition on unseen style reference audio | Tacotron 2 GST · Sally, Born of Darkness evaluation samples | style transfer | Hurt |
| Use unseen-speaker reference audio for posterior style transfer | Flowtron · Sally with RAVDESS surprised and Richard Feynman references | style transfer | Helped |
| Use unseen-speaker reference audio for style transfer | Tacotron 2 GST · Sally with RAVDESS surprised and Richard Feynman references | style transfer | Hurt |
| Translate a Gaussian-mixture dimension associated with pitch height | Flowtron GM-A · LSH speech synthesis | pitch control | Helped |
Closest papers by meaning
- FloWaveNet : A Generative Flow for Raw Audio (ICML 2019 · similarity 0.89)
- DELTA-TTS: Adapting Autoregressive Model into a Diffusion Language Model for Text-to-Speech (NeurIPS 2026 · similarity 0.87)
- SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation (ICML 2025 · similarity 0.87)
- Generative Pre-training for Speech with Flow Matching (ICLR 2024 · similarity 0.86)
- Continuous Autoregressive Modeling with Stochastic Monotonic Alignment for Speech Synthesis (ICLR 2025 · similarity 0.86)
- NaturalSpeech 2: Latent Diffusion Models are Natural and Zero-Shot Speech and Singing Synthesizers (ICLR 2024 · 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.