Updated models source URL in documentation and GUI Reliability enhancement, microphone capture less likely to transition to “Stalled” state and discard the audio
Unclaimed project
Are you a maintainer of Whisper? Claim this project to take control of your public changelog and roadmap.
Changelog
High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model
Updated models source URL in documentation and GUI Reliability enhancement, microphone capture less likely to transition to “Stalled” state and discard the audio
When loading models, adapters can be selected with 0-based index, in addition to the name. Added an API method to decode initial prompt into array of tokens.
Transcribe-File PowerShell command now supports human-readable language names
Better error messages when the library is unable to decode input audio files
New API method to decode audio files from memory buffers. The library should now support concurrent transcribes, when loading multiple models. Added an API method to clone the model, and the corresponding flag in the eGpuModelFlags enumeration. The cloned model uses same GPU and parameters, reuses VRAM buffers with model’s tensors, but it has independent D3D device and context.
The des...
Removed the dependency on cabinet.dll Windows component.
Reportedly, the library now runs on Linux, using Wine 8.3.