Looking for Hook to Capture Progress Bar Status for GUI

[MrEdwards007]

I have created a Streamlit application that enables transcription of media dropped into the GUI.
When I click a button, I start the transcription process. All works as expected but I would like to surface the progress bar information that is available using the following, where verbose=False shows the progress bar

results = model.transcribe(tempTargetPath, fp16=False, word_timestamps = True, verbose=False)

Console output
100%|█████████████████████████████| 370639/370639 [03:47<00:00, 1625.89frames/s]

.
The above progress is going to the console but my program is GUI and does not directly have access to the console information. I did however redirect the console output.
.

Redirected console output
...
98%|#########7| 362652/370639 [03:36<00:04, 1636.68frames/s]
99%|#########8| 365340/370639 [03:38<00:03, 1635.98frames/s]
99%|#########9| 368164/370639 [03:40<00:01, 1601.73frames/s]
100%|##########| 370639/370639 [03:42<00:00, 1570.82frames/s]
100%|##########| 370639/370639 [03:42<00:00, 1668.54frames/s]

# Create a custom Streamlit component to capture stdout and stderr
class StreamlitOutputRedirector:
    def __init__(self):
        self.buffer = ""

    def write(self, text):
        self.buffer += text
        # self.buffer = text.strip() # self.buffer += text
        self.flush()

    def flush(self):
        st.text(self.buffer)
        self.buffer = ""

# Create an instance of the custom StreamlitOutputRedirector
output_redirector = StreamlitOutputRedirector()

# Redirect stdout and stderr to the custom redirector
sys.stdout = output_redirector
sys.stderr = output_redirector

I have done a bit of hack above and redirected stdout and stderr to my program but this is not desirable, as all console output is redirected, not just the progress bar.

I am not redirecting to a file because then I would need to hunt for additional information or errors in the file versus where the user can see the information or errors.

Is there a hook\callback that can be used to capture the progress bar information?

[phineas-pta]

i can only think of 2 possibilities: either edit source code to expose progress bar or write your own transcribe

[glangford]

fyi in case this is helpful, I haven't used it

stqdm : 'stqdm is the simplest way to handle a progress bar in streamlit app'

• https://github.com/Wirg/stqdm

streamlit discussion board on stqdm

• https://discuss.streamlit.io/t/stqdm-a-tqdm-like-progress-bar-for-streamlit/10097/10

[MrEdwards007]

I was able to come up with a solution. its not idea but it is functional. The above is not a completely working program, but the target solution does work. The code is left for the reader to implement based on their needs.

import  streamlit as st
import sys, logging, time

# Create a custom Streamlit component to capture stdout and stderr
class StreamlitOutputRedirector:
    def __init__(self, placeholder):
        self.buffer = ""
        self.placeholder = placeholder

    def write(self, text):
        # Define a keyword to selectively redirect
        keyword_to_redirect = "frames/s]"

        # Check if the keyword is present in the text
        if keyword_to_redirect in text:
            # Redirect text containing the keyword to Streamlit
            # sys.__stdout__.write(text) # Testing

            # Split the input string at each "|"
            split_elements = text.split("|")
            left  = split_elements[0].strip() # contains percentage complete
            right = split_elements[2].strip() # contains "frames/s" information
            newText = left + " | " + right
            self.setText(newText)
        else:
            # Print text without the keyword to the console
            sys.__stdout__.write(text)

    def flush(self):
        # Display the captured output
        # self.placeholder.text(self.buffer)
        # self.buffer = ""

        #Do nothing. The text is already written
        a = 0

    def clear(self):
        # Clear the Streamlit screen by emptying the placeholder
        self.placeholder.empty()
        # self.flush()  # Call flush after writing

    def setText(self, newText):
        # Write content into placeholder
        # self.placeholder.write(newText) # normal writing
        self.placeholder.write(f"###### {newText}") # markdown, write very small

    def replacePlaceholder(self, newPlaceholder):
         # Replace the exising placeholder
         # The desired placeholder may not be initially known
         self.placeholder = newPlaceholder

# Create an empty placeholder for dynamic content
output_placeholder = st.empty()

# Create an instance of the custom StreamlitOutputRedirector
output_redirector = StreamlitOutputRedirector(output_placeholder)

# Redirect stdout and stderr to the custom redirector
sys.stdout = output_redirector
sys.stderr = output_redirector

def transcribeAudio():
    progress = st.sidebar.empty()                   # Create the desired placeholder for progress information
    output_redirector.replacePlaceholder(progress)  # Replace the existing placeholder
    output_redirector.setText("0% complete")        # Set the initial text
    results = model.transcribe(tempTargetPath, fp16=False, word_timestamps = True, verbose=False) 
   
   # ***The progress bar will now automatically capture and update the "progress" place holder***

This doesn't work perfectly.

The updates to the console are not capture perfectly and I don't understand why.
It should capture every update from 0...100 but the output will skip numbers and sometimes may not capture when it arrives at 100%

I love Whisper!

While this is targeted for Streamlit, I suspect this can be modified to work with other GUI interfaces
I dont consider this a closed issue but rather a workaround.

I hope in the future, there will be a hook\callback so this type of workaround is unnecessary.