Back to AI Lab
Generative Models
Research papers, repositories, and articles about generative models
Showing 2 of 2 items
OpenBMB/VoxCPM
Tokenizer-free speech model for multilingual text-to-speech, voice design, and cloning. It skips text tokens entirely and works directly on audio representations. If you build voice products, this shows where foundation TTS models are heading.
23,478
PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation
Flow-matching generators often smooth away sharp temporal patterns. PrismFlow splits the dynamics into a base flow plus expert residuals that specialize on different regimes. If you synthesize time-series data for testing or privacy, this offers more realistic signals without exploding model size.
Junru Zhang, Lang Feng