v0.8.8 v0.8.9
low risk
wxt-dev/wxtSep 26, 2026 → Sep 26, 20260 days between releases
What's New
CPU native updated to 0.4.1, adds demo voice flag and preset profile.
New Features
- Adds `--demo-voice` flag for downloading a public example.
- Adds `--preset windows-3900x` configuration profile.
- Adds `--cpu-fused-attention` flag for F32 attention path.
- Supports GPU server installation via `realtimetts[qwen-server]`.
Improvements
- Reduces mean RTF by ~6.4% on tested Ryzen 3900X hardware.
- Eliminates artificial rebuffering at contiguous audio boundaries.
- Fixes streaming playback issues in Voice Studio.
Bug Fixes
- Resolves standalone synthesis failures for specific language codes.
- Fixes streaming playback issues in Voice Studio.
- Eliminates artificial rebuffering at contiguous audio boundaries.
Upgrade Notes
Update CPU native dependency to 0.4.1. Note that floating-point evaluation order changes may affect seeded audio results. New flags `--demo-voice` and `--preset windows-3900x` are available for easier configuration.
The Two Releases
v0.8.8
RealtimeTTS 0.8.8
Sep 26, 2026· 25 days ago
This release introduces optional CPU decoder overlap to improve generation efficiency by allowing acoustic-code generation to continue while the previous chunk is decoded.
Full release notesv0.8.9
RealtimeTTS 0.8.9
Sep 26, 2026· 25 days ago
This release simplifies the installation and use of Qwen servers while fixing streaming playback issues and standalone synthesis failures.
Full release notes