GPU execution requires the NVIDIA libraries cuBLAS 11.x and cuDNN 8.x to be installed on the system. Please refer to the CTranslate2 documentation.
If executing whisper-ctranslate2 gives errors like:
Could not load library cudnn_ops_infer64_8.dll. Error code 126
Make sure that the environment variable LD_LIBRARY_PATH includes the path where your libraries are installed.
Some audio files can trigger the "temperature fallback" which is based on random sampling. So it is expected that the transcription is different each time. See openai/whisper#81 for more discussion about this.
If you execute whisper-ctranslate2 with or without the Vad filter options it produces different transcriptions.
This is expected since VAD will change the model input.
If you have little memory the program can finish without showing any message.
Trying using a smaller model or a shorter file and watch out your available memory.
Make sure that your terminal has permissions to access the microphone. In macOS, go to Security & Privacy -> Privacy -> Microphone and make sure that the Terminal application has permissions.
Additionally, you can try with a lower volume threshold using --live_volume_threshold argument.
This is a sounddevice library dependency needed for the live transcription functionality. If you are running Ubuntu, you can install the necessary library by running:
sudo apt-get install libportaudio2