ProgressBar handler

[nicko27]

Hi,
I want to use your script with a progressbar but segments doesn't implement len. Is it possible to solve this ? Please

Bonjour d'Evreux (27)

Thanks

[guillaumekln]

Hello,

The transcribe method returns a structure AudioInfo which contains the duration of the audio file. It can be used to implement the progress bar. For example:

from tqdm import tqdm
from faster_whisper import WhisperModel

model = WhisperModel("tiny")
segments, info = model.transcribe("audio.mp3")

total_duration = round(info.duration, 2)  # Same precision as the Whisper timestamps.

with tqdm(total=total_duration, unit=" seconds") as pbar:
    for segment in segments:
        segment_duration = segment.end - segment.start
        pbar.update(segment_duration)

[nicko27]

Thanks

[phineas-pta]

just a minor enhancement, when using SileroVAD, not all segments are continuous of others

from tqdm import tqdm
from faster_whisper import WhisperModel

model = WhisperModel("tiny")
segments, info = model.transcribe("audio.mp3", vad_filter=True)

total_duration = round(info.duration, 2)  # Same precision as the Whisper timestamps.
timestamps = 0.0  # to get the current segments

with tqdm(total=total_duration, unit=" audio seconds") as pbar:
    for segment in segments:
        pbar.update(segment.end - timestamps)
        timestamps = segment.end
    if timestamps < info.duration: # silence at the end of the audio
        pbar.update(info.duration - timestamps)

[Purfview]

How to redirect tqdm's output to print()?

[phineas-pta]

what do you mean? you want to print(tqdm(...)) ?

[Purfview]

I want print() showing a progress bar instead of tqdm().

PS
I think I figured it out, need to test it.

[phineas-pta]

why ? tqdm can handle console progress bar

[Purfview]

@phineas-pta But it can't handle console's title bar, that's where I want to see progress bar.

[phineas-pta]

is there any library to do that ?

[Purfview]

is there any library to do that ?

You don't need lib for that. My problem was solved, but I saw various glitches from tqdm and I think the progress bar examples posted above doesn't work properly.
Today I'll try to debug that and will post a solution.

[Purfview]

Examples above are inaccurate because there tqdm process only a first segment from chunk and the rest are ignored.
Here is my accurate implementation + progress bar in Windows console's title:

import time
import io
from threading import Thread
from time import sleep
from os import system
from tqdm import tqdm
from faster_whisper import WhisperModel


model = WhisperModel("tiny")
segments, info = model.transcribe("audio.mp3")


def pbar_delayed():       # to get last timestamp from chunk
    global timestamp_prev
    sleep(set_delay)      # wait for whole chunk to be iterated
    pbar.update(timestamp_last - timestamp_prev)
    timestamp_prev = timestamp_last
    system("title " + capture.getvalue().splitlines()[-1].replace("|","^|").replace("<","^<").replace("?","0"))
    print(capture.getvalue().splitlines()[-1])

total_dur = round(info.duration)
td_len = str(len(str(total_dur)))      # get length for n_fmt
global timestamp_prev, timestamp_last
timestamp_prev = 0                     # last timestamp in previous chunk
timestamp_last = 0                     # current timestamp
capture = io.StringIO()                # capture progress bars from tqdm
last_burst = 0.0                       # time of last iteration burst aka chunk
set_delay = 0.1                        # max time it takes to iterate chunk & minimum time between chunks.

s_time = time.time()
print("")
bar_f = "{percentage:3.0f}% | {n_fmt:>"+td_len+"}/{total_fmt} | {elapsed}<<{remaining} | {rate_noinv_fmt}"

with tqdm(file=capture, total=total_dur, unit=" audio seconds", smoothing=0.00001, bar_format=bar_f) as pbar:
    for segment in segments:
        timestamp_last = round(segment.end)
        time_now = time.time()
        if time_now - last_burst > set_delay:  # catch new chunk
            last_burst = time_now
            Thread(target=pbar_delayed, daemon=False).start()
    sleep(set_delay + 0.3) # wait for the last pbar_delayed to finish
    if timestamp_last < total_dur: # silence at the end of the audio
        pbar.update(total_dur - timestamp_last)
        system("title " + capture.getvalue().splitlines()[-1].replace("|","^|").replace("<","^<").replace("?","0"))
        print(capture.getvalue().splitlines()[-1])

print("\n\nFinished! Speed: %s audio seconds/s" % round(info.duration / ((time.time() - s_time)), 2))

EDIT:

To get progress in Linux terminal's title try to replace system(.... lines with this:
print('\33]0;'+ capture.getvalue().splitlines()[-1] +'\a', end='', flush=True)