Vibe-coding a Transcriber

I asked Qwen to make a script that makes a transcript for a video, based on an mp3 file. I was careful to tell it to only use a limited set of libraries but this script still required three libraries, pydub, speech_recognition and tqdm, the last one being only for the fancy progress-bar. You can find it here.

Funily enough it stumbled on this part of the code (this is an example of the problem):

import sys
from tqdm import tqdm
import time

size = 10

with tqdm(total=size, desc="Progress", unit="segment", ncols=100, file=sys.stdout) as progress:
    for _ in range(size):
        time.sleep(0.5)
        progress.update(1)

Which shows several progress bars. One per line, instead of a single one like it suggested to see if TQDM was working:

import sys
from tqdm import tqdm
import time

for i in tqdm(range(10), desc="Test Progress Bar", file=sys.stdout):
    time.sleep(0.5)
    # sys.stdout.flush()

Which reinforces my belief that, by themselves, AIs/LLMs are fast but not very smart.

As ColdFusion warned, you should review the transcript that it produced as it makes mistakes.

I later made a video on this, watch it here.

Prompt, in case you’re wondering:

Modify and Remove from the following script to only show one progress bar. Only output the corrected code.

import sys
from tqdm import tqdm
import time

size = 3

with tqdm(total=size, desc="Progress", unit="segment", ncols=100, file=sys.stdout) as progress:
    for _ in range(size):
        time.sleep(0.5)
        progress.update(1)

PS: I updated the example to make it easier to copy+paste.