shirayu / whispering

Streaming transcriber with whisper
MIT License
685 stars 53 forks source link

whispering: error: argument --language: invalid choice: 'multi' #21

Closed anthnyprschka closed 1 year ago

anthnyprschka commented 1 year ago

When using whispering --language multi --model large -n 90 --allow-padding --host 0.0.0.0 --port 8000

shirayu commented 1 year ago

Are you using the latest version? Please update whispering.

Additionally --allow-padding and -n no longer make sense for servers. Use them with clients.

anthnyprschka commented 1 year ago

Thanks, I updated and run server with

whispering --language multi --model tiny --host 0.0.0.0 --port 8000

and client with

whispering --host 127.0.0.1 --port 8000 --mode client -n 30 --allow-padding

Connection is established but nothing transcribed.

I remember it used to work with

whispering --language en --model tiny --host 0.0.0.0 --port 8000

and

whispering --host 127.0.0.1 --port 8000 --mode client.

These are the logs in server:

[2022-10-07 15:48:17,585] transcriber._set_dtype:35 WARNING -> FP16 is not supported on CPU; using FP32 instead
Using cache found in C:\Users\antho/.cache\torch\hub\snakers4_silero-vad_master
C:\Users\antho\dev\voicely\venv\lib\site-packages\torchaudio\backend\utils.py:62: UserWarning: No audio backend is available.
  warnings.warn("No audio backend is available.")
[2022-10-07 15:48:18,306] serve.serve_with_websocket:71 INFO -> Serve at 0.0.0.0:8000
[2022-10-07 15:48:18,307] serve.serve_with_websocket:72 INFO -> Make secure with your responsibility!
[2022-10-07 15:48:18,316] server.wrap:707 INFO -> server listening on 0.0.0.0:8000
[2022-10-07 15:48:32,020] server.handshake:642 INFO -> connection open
C:\Users\antho\dev\voicely\venv\lib\site-packages\whispering\vad.py:50: UserWarning: The given NumPy array is not writable, and PyTorch does not support non-writable tensors. This means writing to this tensor will result in undefined behavior. You may want to copy the array to protect its data or make it writable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  ..\torch\csrc\utils\tensor_numpy.cpp:178.)
  torch.from_numpy(audio[start:end]),
[2022-10-07 15:52:18,088] server.handler:260 INFO -> connection closed
[2022-10-07 15:52:20,835] server._close:755 INFO -> server closing
[2022-10-07 15:52:20,837] server._close:795 INFO -> server closed
shirayu commented 1 year ago

As noted in #23, I noticed that unfortunately multilingual tokenizer is not supposed in Whisper. So, I removed the feature.

Thank you for the report!

anthnyprschka commented 1 year ago

@shirayu is this functioning currently? because i am not getting any transcriptions even with the old commands I sent you above

anthnyprschka commented 1 year ago

it's very confusing. Multilingual works for me now, but I had to set patience to 0.0

shirayu commented 1 year ago

--language multi is deprecated because that is equivalent to --language en.

If you still have a trouble, please open another issue. Thank you.