Create a launch.json file in .vscode

{
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Streamlit",
            "type": "debugpy",
            "request": "launch",
            "module": "streamlit",
            "args": [
                "run",
                "${file}",
                "--server.port",
                "28100"
            ]
        }
    ]
}

Using Commandline and Python

Option 1: Transcribe with Whisper (official)

pip install git+https://github.com/openai/whisper.git

Then run:

whisper demo.mp4 --model medium --language auto --output_format txt

Optional flags:

• --output_format srt (for subtitles)
• --language en (to skip language detection)

Create a shell script to transcribe multiple files

#!/usr/bin/env bash

# Loop through all arguments (filenames)
for VIDEO in "$@"; do
    echo "Transcribing: $VIDEO"
    whisper "$VIDEO" --model medium --language en --output_format txt
done

Option 2: Transcribe with faster-whisper (much faster on CPU or GPU)

pip install faster-whisper

Run via Python:

import sys
from faster_whisper import WhisperModel

VIDEO=sys.arg[1]

model = WhisperModel("medium", device="cpu")  # or "cuda" for GPU
segments, info = model.transcribe(VIDEO)

with open("transcript.txt", "w") as f:
    for s in segments:
        f.write(f"{s.start:.2f} --> {s.end:.2f}: {s.text.strip()}\n")

Optional: Convert to audio (if needed)

If transcription fails or is slow, extract audio first:

ffmpeg -i your_video.mp4 -ar 16000 -ac 1 -c:a pcm_s16le audio.wav

Then transcribe audio.wav instead.

Problems with numpy

Error when running whisper: A module that was compiled using NumPy 1.x cannot be run in NumPy 2.2.6 as it may crash.

Reason:

The error you’re seeing comes from an incompatibility between NumPy 2.x and some Whisper dependencies that were compiled against NumPy 1.x.

Solution: Downgrade NumPy

You can fix this by downgrading NumPy to version 1.x:

pip install "numpy<2"

Then run Whisper again:

whisper demo.mp4 --model medium --language auto --output_format txt

Hide warning UserWarning: FP16 is not supported on CPU; using FP32 instead

Option 1: Suppress all Python warnings (quick + global)

In your terminal or script, set the environment variable:

PYTHONWARNINGS="ignore" whisper dmeo.mp4 --model medium<br></code>

Or, in Python code:

import warnings

warnings.filterwarnings("ignore")

Option 2: Suppress only that specific warning

If you’re using faster-whisper in Python and want to filter only that one:

import warnings

warnings.filterwarnings(
    "ignore",
    message="FP16 is not supported on CPU; using FP32 instead"
)

Option 3: Patch the library (if you’re comfortable)

You can find the line in the faster_whisper source code (usually in transcribe.py) that issues the warning and comment it out or remove it:

# warnings.warn("FP16 is not supported on CPU; using FP32 instead")

Not recommended unless you maintain the code.