Frontend
DriftTTS: One-stage Single Step Text-to-Speech via Generative Drifting
arXiv cs.SD0 views
arXiv:2610.22960v1 Announce Type: new Abstract: Text-to-speech models built on score matching and flow matching produce natural speech, but only by calling the decoder many times, which makes synthesis slow. Existing ways to cut the number of calls either distill from a multi-step teacher or add a stage on top of a pretrained model, leaving single-step TTS without a pretrained TTS model unexplored. We question whether pretraining a TTS model is necessary, and propose DriftTTS, which adopts generative drifting, an objective that moves iterative sampling from inference into training. Drifting assumes samples of fixed shape, whereas speech has variable length; we address this issue with segment drifting. On LJSpeech, DriftTTS matches the multi-step quality of Matcha-TTS and RapFlow-TTS in a single decoder call, while being faster. To our knowledge, this is the first application of generative drifting to text-to-speech.
Source：https://arxiv.org/abs/2610.22960PreviousClinically Informed Acoustic-Aware Attribute Modeling for Respiratory Sound ClassificationNextRethinking Magnitude Compression in Spectral Losses for DDSP Parameter Estimation
Related posts
Your age gate shouldn't compare the estimate to 18
0dev.to
I audited 100 public-sector websites in the EU and the US. The regional ranking is noise.
0dev.to
From sitemap.xml to a full audit report
0dev.to
CLAUDE.md best practices: what to include and skip, with a template
0dev.to
The first two model calls ignore the picker
0dev.to
Building an MCP server for app analytics, so Claude can tell me how my apps are doing
0dev.to