Whisper_CPP Speech Recognition Integration
Whisper_CPP is a C++ high-performance implementation of the OpenAI Whisper model. Compared to the native Python implementation, it offers faster inference and lower memory usage.
Supported Models
Whisper_CPP uses GGML format model files (.bin extension). Supported models:
| Model File | Parameters | VRAM | Quality | Recommended For |
|---|---|---|---|---|
| ggml-tiny.bin | 39M | ~1GB | Fair | Quick testing |
| ggml-base.bin | 74M | ~1GB | Good | Everyday simple use |
| ggml-small.bin | 244M | ~2GB | Good | Multi-language recognition |
| ggml-medium.bin | 769M | ~5GB | Very good | High-accuracy recognition |
| ggml-large-v1.bin | 1550M | ~10GB | Good | Large model v1 |
| ggml-large-v2.bin | 1550M | ~10GB | Very good | Large model v2 |
| ggml-large-v3.bin | 1550M | ~10GB | Best | Highest accuracy |
| ggml-large-v3-turbo.bin | 1550M | ~10GB | Balanced speed/quality | Recommended |
Windows
Since v4.08, the Windows version will auto-download the pre-compiled GPU version on first use and extract it to app_dir/whisper-cpp/.
To use a different version, open https://github.com/ggml-org/whisper.cpp/releases/tag/v1.9.1, download an archive labeled Win32 or x64, extract it, and copy all files from Releases/ to app_dir/whisper-cpp/.
macOS or Linux
For these systems, download the corresponding pre-compiled package for your OS. Copy the extracted binaries and dependencies to app_dir/whisper-cpp/, and grant executable permission to the whisper-cli file.
If no pre-compiled package exists, compile locally according to the repository documentation, then copy the build artifacts and shared libraries (.so etc.) to app_dir/whisper-cpp/. Ensure the whisper-cli executable exists.
Downloading Model Files
Models are auto-downloaded when needed from:
- Primary source (outside China): https://huggingface.co/ggerganov/whisper.cpp/tree/main
- Mirror (China): https://hf-mirror.com/ggerganov/whisper.cpp/tree/main
Place the .bin file in app_dir/models/.
Additional Custom Parameters
Create a pyvideotrans.txt text file in the pyVideoTrans root directory (same folder as sp.exe) to add custom parameters.
FAQ
| Issue | Possible Cause | Solution |
|---|---|---|
| whisper-cli.exe won't start | Missing VC++ runtime | Install Microsoft Visual C++ Redistributable |
| Model file not found | Model not downloaded or wrong path | Place .bin file in models/ |
| GPU acceleration not working | CUDA/Vulkan not configured | Install matching CUDA Toolkit version or update GPU driver |
| Very slow recognition | Using CPU version | Download the CUDA-accelerated version and configure correctly |
| Poor recognition results | Model too small or wrong language | Use a larger model, verify language setting |
| Chinese path errors | Path contains Chinese characters | Extract pyVideoTrans to a path without Chinese characters |