Research map / Image generation / Autoregressive and latent generation
P-Flow: A Fast and Data-Efficient Zero-Shot TTS through Speech Prompting
Sungwon Kim, Kevin J. Shih, Rohan Badlani, Joao Felipe Santos, Evelina Bakhturina, Mikyas T. Desta, Rafael Valle, Sungroh Yoon, Bryan Catanzaro
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 fixed-size speaker embedding with speech prompt in text encoder (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.373 -> 0.544 | Improved |
| replace fixed-size speaker embedding with speech prompt in text encoder (ablation) | P-Flow · LibriSpeech test-clean | WER: 2.9 -> 2.6 | No clear effect |
| remove classifier-free guidance (gamma 1 -> 0) (ablation) | P-Flow · LibriSpeech test-clean | WER: 2.6 -> 3.7 | Hurt |
| remove classifier-free guidance (gamma 1 -> 0) (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.544 -> 0.492 | Hurt |
| increase guidance scale from 1 to 2 (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.544 -> 0.546 | No clear effect |
| reduce Euler steps from 10 to 1 (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.544 -> 0.420 | Hurt |
| reduce Euler steps from 10 to 2 (ablation) | P-Flow · LibriSpeech test-clean | WER: 2.6 -> 2.9 | Hurt |
| increase Euler steps from 10 to 20 (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.544 -> 0.540 | No clear effect |
| increase Euler steps from 1 to 20 (ablation) | P-Flow · LibriSpeech | MOS: 3.55 -> 4.14 | Improved |
| use Heun's second-order method with 4 steps instead of 10 Euler steps (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.544 -> 0.552 | Improved |
| use midpoint method with 4 steps instead of 10 Euler steps (ablation) | P-Flow · LibriSpeech test-clean | SECS: 0.544 -> 0.540 | No clear effect |
| P-Flow vs VALL-E in latency | P-Flow · LibriSpeech | inference latency (s): 2.515 -> 0.115 | Improved |
Closest papers by meaning
- Mega-TTS 2: Boosting Prompting Mechanisms for Zero-Shot Speech Synthesis (ICLR 2024 · similarity 0.89)
- Enhancing Zero-Shot Multi-Speaker TTS with Negated Speaker Representations (AAAI 2024 · similarity 0.88)
- StableVC: Style Controllable Zero-Shot Voice Conversion with Conditional Flow Matching (AAAI 2025 · similarity 0.88)
- Generative Pre-training for Speech with Flow Matching (ICLR 2024 · similarity 0.87)
- CLaM-TTS: Improving Neural Codec Language Model for Zero-Shot Text-to-Speech (ICLR 2024 · similarity 0.87)
- Kinetic-Optimal Scheduling with Moment Correction for Metric-Induced Discrete Flow Matching in Zero-Shot Text-to-Speech (NeurIPS 2026 · similarity 0.87)
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.
