Whisper Engine

LLPlayer currently supports two Whisper engines. Each engine is explained below.

whisper.cpp

official: https://github.com/ggerganov/whisper.cpp
using library: https://github.com/sandrohanea/whisper.net

whisper.cpp is an whisper engine that can be used immediately after downloading LLPlayer.
You must download a model beforehand in settings.

Using CUDA or Vulkan or OpenVINO, you can generate subtitles very fast. However, there are currently the following problems.

• inaccurate subtitle timestamps
• serious bug that generates the same subtitle over and over again (hallucination, repeat bug)
  • issue #2

Currently, This engine is not recommended.

faster-whisper (recommended)

official: https://github.com/SYSTRAN/faster-whisper
using CLI: https://github.com/Purfview/whisper-standalone-win

faster-whisper requires a separate engine download in LLPlayer. Please download it from the ASR settings.
The models are automatically downloaded in the first run.

faster-whisper runs fast on CPUs and can generate even faster if CUDA is available.

The subtitle timestamps are very accurate and the operation is very stable with no hallucination bugs.

This engine is recommended.