Progress bar not displaying when using 'print_progress=True'

[SirDaXll]

I noticed that Whisper uses tqdm to show a progress bar when verbose=False. In the whisper.cpp project, there's a function named print_progress for this purpose, which is also present in pywhispercpp inside constants.py under PARAMS_SCHEMA (also referenced in the API documentation).

However, when I try to use this functionality, I don't see any progress bar—only the final transcription result is returned.

I tried using in the command line:

pwcpp copy.mp3 -m medium --output-txt --print_realtime true --print_progress true

And in Python:

from pywhispercpp.model import Model

model = Model('tiny')
segments = model.transcribe('copia.mp3', language='es', print_progress=True, print_realtime=True)

print_realtime works fine and are in the same location as the other

I installed the package from the source, use Windows 11 and Python 1.12.6.

This functionality is currently supported in pywhispercpp or if additional configuration is required?

[absadiki]

Thanks @SirDaXll, for reporting the issue,

I noticed that print_progress now requires a progress callback, likely due to recent changes in whispercpp, that's why you didn't see anything. I tried to add a default callback with a simple message in case print_progress is True, but you can override this by defining your own custom progress_callback:

res = model.transcribe(media="audio.mp3", print_progress=True, progress_callback=lambda progress: print(progress))

Please pull the latest commit and try again. Let me know if you have any further issues.

[SirDaXll]

@absadiki What seems most relevant to me about the output is the time elapsed/time remaining in progress (how output should be: 35%|███▌ | 5588/15901 [00:17<00:31, 326.60frames/s]), although the most important thing is to know that the transcription is progressing, so that's great. If you can figure it out completely, I'd be more grateful than I already am, and if I have time to review it, I'll share it as soon as I can. For now, thanks for the solution!

[absadiki]

@SirDaXll, if you are just looking for a progress bar like effect, you can achieve that easily by using tqdm and the progress_callback I mentioned earlier:

with tqdm(total=100) as pbar:
    res = model.transcribe(media="audio.mp3",
                       print_progress=True,
                       progress_callback=lambda progress: (setattr(pbar, 'n', progress), pbar.refresh()),
                       )

however, whisper.cpp does not provide a progress bar with elapsed/time remaining AFAIK.

[SirDaXll]

@absadiki It is a good solution in the absence of the progress bar, it only works as a reference, since in the absence of having a function that obtains the total frames of the audio and indicates it to tqdm (as Whisper does), the progress will not be precise. I tried the frames thing with pydub, I still didn't get good results, so for the moment I'll stick with what there is.

[SirDaXll]

@absadiki After the update I did not try to execute from the command line, only from a .py file, and now I have an error when receiving the parameters. I tried in 2 computers, with and without venv, just to check.

>pwcpp copy.mp3 -m small -otxt

PyWhisperCpp
A simple Command Line Interface to test the package
Version: 1.3.1.dev12+ga484dc2
====================================================

whisper_init_from_file_with_params_no_state: loading model from 'C:\Users\myuser\AppData\Local\pywhispercpp\pywhispercpp\models\ggml-small.bin'
whisper_init_with_params_no_state: use gpu    = 1
whisper_init_with_params_no_state: flash attn = 0
whisper_init_with_params_no_state: gpu_device = 0
whisper_init_with_params_no_state: dtw        = 0
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51865
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 768
whisper_model_load: n_audio_head  = 12
whisper_model_load: n_audio_layer = 12
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 768
whisper_model_load: n_text_head   = 12
whisper_model_load: n_text_layer  = 12
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 3 (small)
whisper_model_load: adding 1608 extra tokens
whisper_model_load: n_langs       = 99
whisper_model_load:      CPU total size =   487.01 MB
whisper_model_load: model size    =  487.01 MB
whisper_init_state: kv self size  =   56.62 MB
whisper_init_state: kv cross size =   56.62 MB
whisper_init_state: kv pad  size  =    4.72 MB
whisper_init_state: compute buffer (conv)   =   22.41 MB
whisper_init_state: compute buffer (encode) =  280.07 MB
whisper_init_state: compute buffer (cross)  =    6.18 MB
whisper_init_state: compute buffer (decode) =   97.27 MB
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "C:\Python312\Scripts\pwcpp.exe\__main__.py", line 7, in <module>
  File "C:\Users\myuser\AppData\Roaming\Python\Python312\site-packages\pywhispercpp\examples\main.py", line 96, in main
    run(args)
  File "C:\Users\myuser\AppData\Roaming\Python\Python312\site-packages\pywhispercpp\examples\main.py", line 48, in run
    logging.info(f"System info: n_threads = {m.get_params()['n_threads']} | Processors = {args.processors} "
                                             ^^^^^^^^^^^^^^
  File "C:\Users\myuser\AppData\Roaming\Python\Python312\site-packages\pywhispercpp\model.py", line 169, in get_params
    res[param] = getattr(self._params, param)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: Unable to convert function return value to a Python type! The signature was
        (self: _pywhispercpp.whisper_full_params) -> function

[absadiki]

@SirDaXll, it seems that the Python getattr function is having issues with converting the new progress_callback function I added in the previous commit, I think we can ignore it.
Give it a try now; hopefully, it is fixed.

[SirDaXll]

@absadiki With that last commit I found the problem that now some parameters entered by the command prompt did not work (in the .py file there are no problems), so I went into the model.py code to find the problem, locating different log captures in the terminal and see what happens, well I found that it captured the parameters but reset them to certain default values ​​in "_init_model", so I commented a line and the problem was solved. In the pull request I better describe what happened.

def _init_model(self) -> None:
    """
    Private method to initialize the method from the bindings, it will be called automatically from the __init__
    :return:
    """
    logger.info("Initializing the model ...")
    with utils.redirect_stderr(to=self.redirect_whispercpp_logs_to):
        self._ctx = pw.whisper_init_from_file(self.model_path)
        # Don't set the params to the default ones here, as it will overwrite the user's params
        # self._params = pw.whisper_full_default_params(pw.whisper_sampling_strategy.WHISPER_SAMPLING_GREEDY)