only works in cpu mode , but gpu outputs nothing

[yslion]

os: win11 , cuda 11.6 cudnn 8.9 rtx3060 4G

every example in readme i've tried , if no --device , no outputs , if --device cpu it works but slow.

and I've tried the newest version

whisper-ctranslate2-0.1.9
whisper-ctranslate2-0.2.0

[jordimas]

Does it make any difference if you use base, tiny, small or large-v2 models?

[yslion]

@jordimas

large-v2 is still downloading , but the other 3 , shows nothing

(venv) D:\py\whisper-ctranslate2>whisper-ctranslate2 1.mp3 --model medium
Detecting language using up to the first 30 seconds. Use --language to specify the language
Estimating duration from bitrate, this may be inaccurate

(venv) D:\py\whisper-ctranslate2>whisper-ctranslate2 1.mp3 --model base
Detecting language using up to the first 30 seconds. Use --language to specify the language
Estimating duration from bitrate, this may be inaccurate

(venv) D:\py\whisper-ctranslate2>whisper-ctranslate2 1.mp3 --model small
Detecting language using up to the first 30 seconds. Use --language to specify the language

[jordimas]

Can you try an older version like:

pip3 install whisper-ctranslate2==0.1.0

to see if it works? Thanks

[yslion]

Can you try an older version like:

pip3 install whisper-ctranslate2==0.1.0

to see if it works? Thanks

I installed 0.1.0 just now , it said : Could not load library cudnn_cnn_infer64_8.dll. Error code 126 Please make sure cudnn_cnn_infer64_8.dll is in your library path!
then I go to search and try a way : https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#installcuda-windows and down 'ZLIB DLL.' , unzip and copy dll_x64\zlibwapi.dll to windows\system32

it works

I update whisper-ctranslate2 to -0.1.9 , it works too

But there is a problem about live_transcribe , 0.1.0 does not have this command

in 0.1.9 , whisper-ctranslate2 --live_transcribe True --language en or leave '--language' empty or fill with other language , It works at most 6 dialogs and then quit without any message

is it possiable out of memery ? how to get details or some logs

[Zacharie-Jacob]

when you say output, do you mean specifically not saving the files out? Does it display to the termianl, though?

There is this issue about display but no saved files:

#11

[zx3777]

display to the termianl but not save , everytime i use this. I have no choice but to use the original version, which is slow but no problem.

1.0 could save，but when i use --word_timestamps=ture ，the --highlight_words=true will auto turn on，and cant be closed.

[yslion]

@Zacharie-Jacob
no, 'no outputs' I said , means no display about translating , no save files

this is my cmd line
whisper-ctranslate2 1.mp3 --model medium

this is it's output

Detecting language using up to the first 30 seconds. Use --language to specify the language
Estimating duration from bitrate, this may be inaccurate

and then quit without showing anything

[efonte]

The problem is in the faster_whisper package. If you run for example this script it will show you what the problem is.

from faster_whisper import WhisperModel

model_size = "large-v2"
model = WhisperModel(model_size, device="cuda", compute_type="float16")
segments, info = model.transcribe("audio.wav", beam_size=5)
print("Detected language '%s' with probability %f" % (info.language, info.language_probability))
for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

PS C:\> python .\script.py
Could not load library cudnn_cnn_infer64_8.dll. Error code 126
Please make sure cudnn_cnn_infer64_8.dll is in your library path!

The error has been solved by installing zlib:
https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-zlib-windows

Maybe it would be good to show the error in whisper-ctranslate2 since it hides it.

[yslion]

The problem is in the faster_whisper package. If you run for example this script it will show you what the problem is.

from faster_whisper import WhisperModel

model_size = "large-v2"
model = WhisperModel(model_size, device="cuda", compute_type="float16")
segments, info = model.transcribe("audio.wav", beam_size=5)
print("Detected language '%s' with probability %f" % (info.language, info.language_probability))
for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

PS C:\> python .\script.py
Could not load library cudnn_cnn_infer64_8.dll. Error code 126
Please make sure cudnn_cnn_infer64_8.dll is in your library path!

The error has been solved by installing zlib: https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-zlib-windows

Maybe it would be good to show the error in whisper-ctranslate2 since it hides it.

thank you for your reply , I ve fix the zlib issue (see above ,line 5 ) .

the matter is , live_transcribe , print out 'Transcribing..' and the quit without any outputs

(venv) D:\py\whisper-ctranslate2>whisper-ctranslate2 --live_transcribe True --language en  --model tiny 
Listening.. (Ctrl+C to Quit)
...............................................................................................
Transcribing..
...........
(venv) D:\py\whisper-ctranslate2>

and @jordimas

[jordimas]

This has been fixed with version 0.26