ASR: disable progress-bar for transcription?

[domcross]

Hello all,

probably I am missing something obvious - is it possible to disable the progress-bar (tqdm) for the transcribe-method of an asr-model?

Thanks for any suggestions in advance.

Dominik

[titu1994]

Not particularly, is there a usecase for removing the progress bar?

[domcross]

I want to run a custom evaluation script with my own pre/post-processing where I have set batchsize to "1".
Currently I have two progress bars (one from asr-model.transcribe and one from my sourrounding processing loop) which overwrite each other...

[titu1994]

Removing the progress bar will be tricky. Nested tqdm should work fine though. If we do need to support this it will be a change in too many files and make the transcribe() method have to accept another argument verbose=True which would not be great since in nearly all cases you want that to be true

[prakhark28]

Adding a verbose argument makes sense as almost every time one uses the transcribe function, it is inside an outer loop with tqdm. The transcribe bar messes up the output logs.

[edStorey]

I have also had the same problem as @domcross, I am looping through several files using the transcribe function with a batch size of 1 similar to the above.

I used this fix codereview.stackexchange.com/a/25471 to get around the issue by running the transcribe function in a with statement with their NoStdStreams class this stopped the progress bar from appearing. This was not ideal however, functionality to disable progress bars would be appreciated in future nemo releases.

[prakhark28]

for audio in tqdm(data):    
    with NoStdStreams():
        logits.append(asr_model.transcribe([audio['audio_filepath']], logprobs=True)[0])

This returns a tqdm and transcribe progress bar alternately. Can you help me mute the transcribe outputs

[edStorey]

class NoStdStreams(object):
        def __init__(self,stdout = None, stderr = None):
            self.devnull = open(os.devnull,'w')
            self._stdout = stdout or self.devnull or sys.stdout
            self._stderr = stderr or self.devnull or sys.stderr

        def __enter__(self):
            self.old_stdout, self.old_stderr = sys.stdout, sys.stderr
            self.old_stdout.flush(); self.old_stderr.flush()
            sys.stdout, sys.stderr = self._stdout, self._stderr

        def __exit__(self, exc_type, exc_value, traceback):
            self._stdout.flush(); self._stderr.flush()
            sys.stdout = self.old_stdout
            sys.stderr = self.old_stderr
            self.devnull.close() 

That is the NoStdStreams class I used ^^

with NoStdStreams() : 
    transcription = asr_model.cuda().transcribe(paths2audio_files=file_path)

And that is my implementation, performing the transcription on gpu with cuda().

Hope that helps!

[evilc3]

I used these lines to disable tqdm completely

`from tqdm import tqdm

from functools import partialmethod

tqdm.init = partialmethod(tqdm.init, disable=True)
`

[tylercasper]

This is an old topic, but when streaming, it is very desirable to have the progress and inference rate output be suppressed. With an 80 ms lookahead and 80 ms encoder step length, this output is printed excessively.