# Version history and release differences

## Which version does this documentation describe?

The public source inspected for these pages identifies itself as 2.1.0. Check your installed version under **Other → Update Ilaria RVC**. Use **All changes so far** there to browse the release-by-release notes even offline.

## All releases

| Release | Highlights |
| --- | --- |
| 2.1.0 | Added the website button, live dashboard metrics for the active run, and better startup and refresh behavior in embedded TensorBoard. |
| 2.0.0 | Introduced the Training Dashboard, embedded TensorBoard, diagnostic reports and 12-second Dataset Checker previews. Optional embedders became individual downloads. |
| 1.9.3 | Introduced advanced audio processing, five inference export formats, configurable embedders and custom HuBERT installation. |
| 1.9.2 | Added public Google Drive and MEGA links to the voice model downloader. |
| 1.9.1 | Added sample-rate checks when reusing an experiment name, avoiding a training crash on incompatible configurations. |
| 1.9.0 | Added FCPE, streamed model imports with cancellation, a reorganized Other tab and expandable offline release notes. |
| 1.8.1 | Updated completion and test sounds while retaining the Windows error sound. |
| 1.8.0 | Introduced the hidden-console launcher, UI restart control, Audio Analyzer and a shared folder layout for models and their indexes. |
| 1.7.2 | Fixed conversion without an index and named Single and Batch output files after the selected voice. |
| 1.7.1 | Suppressed a specific Windows connection-close traceback while preserving other error messages. |
| 1.7.0 | Added Edge and offline Windows TTS, waveform players and background preparation for pitch extraction and supported default models. |
| 1.6.0 | Added the persistent training queue, exact step targets, RefineGAN and Spin-v2 options, paired pretrained downloads, more pitch extractors and ZIP export. |
| 1.5.2 | Improved duplicate-launch diagnostics and checked dataset availability before each training stage. |
| 1.5.1 | Blocked preprocessing while a dataset import was active or the dataset path was unusable. Added the offline change history panel. |
| 1.5.0 | Introduced drag-and-drop dataset import and management, the PC training advisor and optional completion and error alerts. |
| 1.4.0 | Added training previews, resumable Stop and save, and the Dataset Checker. |
| 1.3.0 | Validated the update path from 1.2, showed the installed version in the console and prepared a smaller Windows package. |
| 1.2.0 | Added the built-in updater with verification and recovery, FP32/FP16 training selection, and Model information. |
| 1.1 Beta | Introduced the themed interface, numbered training workflow, drag-and-drop conversion, automatic index matching, model imports and merging, and the app glossary. |

> Generic HuBERT checkpoints are not recommended as the default encoder for voice conversion. Their speech prediction training does not explicitly remove speaker identity, so the features can carry source timbre into the output. For RVC, use the encoder expected by the model, such as ContentVec when the voice was trained with it.

If a button or backend differs on your build, compare its release notes before assuming a failed install. The current UI snapshot documents Single, Batch and TTS conversion.