Skip to content

AI-RVC v1.6.0

Latest

Choose a tag to compare

@mason369 mason369 released this 23 Oct 18:45

这次把 TelkNet 同源的翻唱更新同步到了独立版。

2026-10-26 同版本修复:解决部分电脑启动时 Numba 报 cannot cache function / no locator available 的打包问题,补齐实际成品的混响、重采样、混音和中英文启动检查。版本号仍为 1.6.0。请重新下载所选包的全部分卷,解压到新目录;不要混用旧版分卷。

同时修正前端开发依赖误入便携包的问题,播放器继续完整内置,可离线使用。

修复语言切换或刷新后出现“无效翻唱模式”、动态表格和播放器提示残留旧语言的问题,中英文均已补充往返操作检查。

  • 已有干声可以直接换声,结果保留输入和转换后人声两轨,方便试听对比。
  • 改善转音和长歌曲拼接,保留模型原生精度;修复双索引角色包、首次模型准备和官方 UVR5 路线的问题。
  • 中英文界面、多轨播放和原始 WAV 下载同步更新。提供 Windows / Linux 的 CPU、GPU 便携包,以及 CPU、CUDA 容器镜像。

下载所选版本的全部分卷,放在同一目录后解压;可用附带的 SHA-256 文件检查完整性。

English

This release brings TelkNet's shared cover updates to the standalone app.

2026-10-26 same-version fix: resolve the packaging issue that causes Numba's cannot cache function / no locator available on some computers, with actual frozen reverb, resampling, mixing and bilingual startup checks. The version remains 1.6.0. Download every volume for your chosen package again and extract into a new directory; do not mix old and rebuilt volumes.

Also stop bundling frontend development dependencies. The complete player remains included for offline use.

Fix invalid conversion-mode errors and stale dynamic table/player translations after changing language or reloading. Both Chinese and English now have round-trip interaction checks.

  • Convert existing dry vocals directly, with separate input and converted tracks for comparison.
  • Improve pitch transitions and long-song joins while preserving native model precision. Fix paired character indices, first-time model preparation and the official UVR5 route.
  • Update both interface languages, multitrack playback and original WAV downloads. Windows/Linux CPU and GPU portable packages and CPU/CUDA container images are available.

Download every volume for your chosen package into one folder before extracting. The included SHA-256 files let you check that the download is complete.

便携包重新打包 / Portable rebuild: e94f73bf1768.