0.2.5

QuentinFuxa/WhisperLiveKit0.2.5Aug 13, 2025by QuentinFuxa

AI Summary

This release (0.2.5) of WhisperLiveKit introduces significant backend architecture improvements including multi-user support for shared Whisper model instances, enhanced performance monitoring with faster lag metric updates, and reduced hallucinations during silent periods. The frontend also receives notable enhancements with dark theme support, three distinct silence indicators, and improved UX during transcription.

Key Highlights

Refactored SimulStreaming backend to allow multiple users to share the same Whisper model instance
Enhanced performance monitoring with lag metrics updating every 0.1 seconds
Added dark theme support for the frontend
Reduced false transcriptions (hallucinations) during silent periods
Migrated to pyproject.toml for modern Python packaging

New Features

Multi-user backend architecture - multiple users can share the same Whisper model instance
Enhanced performance monitoring - lag metrics update every 0.1 seconds independently of token emission frequency
Reduced hallucinations - SimulStreaming is less likely to generate false transcriptions during silent periods
Three distinct silence indicators: model-detected silences ([BLANK_AUDIO]), token emission gaps, and end-of-transcription silences
Dark theme support
Screen wake lock - screen no longer goes to sleep while transcribing
Auto-scroll - automatically scrolls to the latest transcription text

Full Release Notes

**Build & Dependencies**
- **Migrated to `pyproject.toml`** - Replaced `setup.py` with PEP-recommended packaging [bda72b8](https://github.com/QuentinFuxa/WhisperLiveKit/commit/bda72b8bc0bc22fa06584bd3f3dc1f3c3615a104)
- **Removed NumPy version constraint** - No longer restricted to `numpy < 2.0.0` [197293e](https://github.com/QuentinFuxa/WhisperLiveKit/commit/197293e25e91b2b9aea514d5f2dae7c99670e8ee)

**Backend Architecture**
- **Refactored SimulStreaming backend separation** - Improved architecture to allow multiple users to share the same backend Whisper model instance https://github.com/QuentinFuxa/WhisperLiveKit/commit/d098af31850a616a93d3e2eb134f3cab3695024f https://github.com/QuentinFuxa/WhisperLiveKit/commit/197293e25e91b2b9aea514d5f2dae7c99670e8ee
- **Enhanced performance monitoring** - Lag metrics now update every 0.1 seconds and are independent of token emission frequency https://github.com/QuentinFuxa/WhisperLiveKit/commit/2bbdc70187ce6ada70f2ed569713bcc3fae9c227
- **Reduced hallucinations** - SimulStreaming is now less likely to generate false transcriptions during silent periods https://github.com/QuentinFuxa/WhisperLiveKit/commit/87b9ed6ecdf2753b6060216b813e3008a75ef388

**Frontend Improvements**
- **Enhanced silence indicators** - Now displays three distinct types of silences:
  - Model-detected silences (`[BLANK_AUDIO]`)
  - Token emission gaps
  - End-of-transcription silences  
  https://github.com/QuentinFuxa/WhisperLiveKit/commit/38b4ebe8ba5621ba6a15f4e053c5f6a996093c51
- **Dark theme support** - Added dark mode https://github.com/QuentinFuxa/WhisperLiveKit/commit/4e56130a40af118c7d7d0c651e7d8f6634acaad9
- **Improved UX during transcription** by @davidgumberg 
  - Screen no longer goes to sleep while transcribing https://github.com/QuentinFuxa/WhisperLiveKit/commit/7f93c4b9787d2ccca19ee27ac521438a90748fbe
  - Auto-scroll to latest transcription text https://github.com/QuentinFuxa/WhisperLiveKit/commit/3b96fb8776de399f1f94700d3636888513d7a1bb

<img width="726" height="349" alt="Screenshot 2025-08-11 at 17 53 50" src="https://github.com/user-attachments/assets/8f9bb573-f7db-45e0-96ac-2213fce52505" />