Could not locate cudnn_ops_infer64_8.dll.

[wo198777]

PS C:\Users\cheng> & C:/Users/cheng/anaconda3/envs/daily/python.exe d:/code/spy/实时语音转文字.py

Wait until it says '开始'

\ transcribingCould not locate cudnn_ops_infer64_8.dll. Please make sure it is in your library path!
RealTimeSTT: root - ERROR - Error during transcription:
Traceback (most recent call last):
File "C:\Users\cheng\anaconda3\envs\daily\Lib\multiprocessing\connection.py", line 328, in _recv_bytes
nread, err = ov.GetOverlappedResult(True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
BrokenPipeError: [WinError 109] 管道已结束。

[KoljaB]

Means in most cases faster_whisper isn’t compatible with cuDNN >9 and PyTorch >2.4.

How to fix:

Option 1: Downgrade

• Lower cuDNN below version 9, or
• Lower PyTorch below version 2.4.

Option 2: Preserve Versions
Keep cuDNN 9 and PyTorch ≥2.4 by doing this:

1. Create a fresh virtual environment.
2. Install an older torch version (e.g., 2.3.1 or 2.1.2 with CUDA).
3. In that venv, locate cudnn_cnn_infer64_8.dll and cudnn_ops_infer64_8.dll under lib/site-packages/torch/lib.
4. Copy these files to the corresponding folder in your current torch installation or to system32.

Option 3: Use DLLs from cuDNN and Toolkit
Alternatively, copy all DLLs from cuDNN and cublasLt64_11.dll from the GPU Computing Toolkit into your ctranslate2 package directory. For example, if you use a venv, it’s usually at \venv\Lib\site-packages\ctranslate2. Paths will vary with Conda or other setups (for more details, see this GitHub comment).