How to Transcribe Video to Text Using OpenAI Whisper

brew install ffmpeg

sudo apt update
sudo apt install ffmpeg

ffmpeg -version

pip install openai-whisper

pip install openai-whisper torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

ffmpeg -i interview.mp4 -vn -acodec libmp3lame -q:a 2 interview.mp3

ffmpeg -i interview.mp4 -vn -ar 16000 -ac 1 -b:a 64k interview.mp3

whisper interview.mp3 --language English --model medium

whisper interview.mp3 --language English --model medium --output_format txt

ffmpeg -i user_feedback_session.mov -vn -ar 16000 -ac 1 feedback.mp3

whisper feedback.mp3 --language English --model medium --output_format txt

So the first thing about these signals before the interval.
Yes, so you set a 25-second interval and before the interval
you get 4 audio signals, let's say, the main signal with an
accent and then 4 weaker ones again. So basically if you don't
arrive exactly on the signal, you know that you need to speed
up by two seconds or slow down by two seconds because the sound
tells you that.

#!/usr/bin/env python3
"""
Video to Text Transcription Pipeline
Uses FFmpeg + OpenAI Whisper
"""

import subprocess
import whisper
from pathlib import Path


def extract_audio(video_path: str, audio_path: str) -> None:
    """Extract audio from video using FFmpeg."""
    cmd = [
        "ffmpeg", "-i", video_path,
        "-vn", "-ar", "16000", "-ac", "1",
        "-b:a", "64k", "-y",
        audio_path
    ]
    subprocess.run(cmd, check=True, capture_output=True)
    print(f"Audio extracted: {audio_path}")


def transcribe_audio(
    audio_path: str,
    output_path: str,
    model_name: str = "medium",
    language: str = "en"
) -> str:
    """Transcribe audio using Whisper."""
    print(f"Loading model {model_name}...")
    model = whisper.load_model(model_name)

    print("Transcribing...")
    result = model.transcribe(audio_path, language=language)

    text = result["text"]

    with open(output_path, "w", encoding="utf-8") as f:
        f.write(text)

    print(f"Transcription saved: {output_path}")
    return text


def main():
    # Configuration
    video_file = "meeting_recording.mp4"
    audio_file = "temp_audio.mp3"
    output_file = "meeting_transcript.txt"

    # Pipeline
    extract_audio(video_file, audio_file)
    text = transcribe_audio(audio_file, output_file)

    # Cleanup temp file
    Path(audio_file).unlink()

    print(f"\nTranscription preview:\n{text[:500]}...")


if __name__ == "__main__":
    main()

python transcribe.py

import whisper

model = whisper.load_model("medium")
result = model.transcribe("podcast_episode.mp3", language="en")

# Print segments with timestamps
for segment in result["segments"]:
    start = segment["start"]
    end = segment["end"]
    text = segment["text"].strip()
    print(f"[{start:.1f}s - {end:.1f}s] {text}")

[0.0s - 4.2s] Welcome to the show. Today we're talking about...
[4.2s - 8.7s] ...building hardware products for athletes.
[8.7s - 15.3s] Our guest has been working on a tempo trainer device.

import whisper

model = whisper.load_model("medium")

# Auto-detect language
result = model.transcribe("unknown_language.mp3")

print(f"Detected language: {result['language']}")
print(f"Text: {result['text']}")

pip install mlx-whisper

import mlx_whisper

result = mlx_whisper.transcribe(
    "audio.mp3",
    path_or_hf_repo="mlx-community/whisper-medium-mlx"
)

ffmpeg -i long_recording.mp3 -f segment -segment_time 1800 -c copy chunk_%03d.mp3

result = model.transcribe(
    "audio.mp3",
    language="en",
    initial_prompt="This is a conversation about swimming training and interval timers."
)