How I Saved 5 Hours by Building My Own AI Transcriber Locally (You Can Do the Same)

3 minute read

I recently had to write a report from a 1-hour-long seminar. I had slides and a recording—everything I needed. Still, the thought of playing it back, pausing, rewinding, and writing notes felt like a 5–6 hour chore.

But instead of grinding through it, I spent just one hour.
💡 30 minutes to build a transcription script
💡 30 minutes to edit the report using the transcript

This is where local AI automation shines—and if you’re a researcher, engineer, or obsessive note-taker like me, this post might just change your workflow.

Why I Built a Local Transcription Setup

I’ve always loved automating repetitive tasks. It’s not just about saving time—it’s about doing work once and reusing it forever. As someone who takes notes extensively in Obsidian, I regularly capture content from YouTube videos, webinars, and Zoom lectures.

Why?

Because when I need to revisit a specific detail, I don’t want to watch an entire video again. I want to search my notes instantly. That’s where transcription comes in.

Why I Don’t Use Online Services

There are many transcription tools online, but most have one or more of the following issues:

Limited to 30-minute videos
Slow processing or queue times
Paid subscriptions or hidden costs
Privacy risks—especially critical when dealing with internal or corporate resources

Most companies don’t allow uploading internal meetings to third-party servers. Local transcription is the only option. I’ve even used this for missed seminars—transcribing them and extracting key ideas in minutes.

How to Set Up Local Video Transcription with Whisper

Let’s walk through the process to set up Whisper, OpenAI’s powerful transcription model, locally on your machine.

Step 1: Create a Python Environment

First, create and activate a Conda environment:

conda create --name transcribe python=3.10
conda activate transcribe

Step 2: Install Required Packages

Install Whisper and necessary tools:

pip install openai-whisper
pip install torch torchvision torchaudio
conda install -c conda-forge ffmpeg

Step 3: Verify Installation

Before using Whisper, make sure it’s installed correctly:

import whisper
import subprocess

try:
    subprocess.run(['ffmpeg', '-version'], capture_output=True, check=True)
    print("FFmpeg accessible from Python")
except Exception as e:
    print(f"FFmpeg issue: {e}")

try:
    model = whisper.load_model("tiny")
    print("✅ Whisper model loaded successfully")
except Exception as e:
    print(f"Whisper issue: {e}")

Expected Output:

FFmpeg accessible from Python 
Whisper model loaded successfully

📝 Transcribe Your Video or Audio

Here’s the script I use to transcribe any video file and automatically save the output in a .txt file.

import whisper
from pathlib import Path

VIDEO_PATH = r"..\Obsidian\01 Obsidian Getting Started\00 Obsidian Getting Started V2.mov"
MODEL_SIZE = "base"
LANGUAGE = None  # Set to "en", or leave None for auto-detect

def transcribe_video():
    model = whisper.load_model(MODEL_SIZE)
    print(f"Transcribing: {Path(VIDEO_PATH).name}")
    
    result = model.transcribe(VIDEO_PATH, language=LANGUAGE)
    
    output_path = Path(VIDEO_PATH).parent / f"{Path(VIDEO_PATH).stem}_transcription.txt"
    with open(output_path, 'w', encoding='utf-8') as f:
        f.write(result["text"])

    print(f"Saved to: {output_path}")
    print("\nPreview:\n", result["text"][:300], "...")
    return result["text"]

if __name__ == "__main__":
    transcribe_video()

My Workflow: Whisper + Obsidian = Instant Knowledge

I save each transcription in Obsidian, link it with related notes, and build a searchable, permanent second brain. Now if someone asks what was discussed in a 90-minute session last year, I have the answer in seconds.

Final Thoughts

This tiny setup has saved me countless hours—and it cost me nothing. No waiting. No cloud. Just offline AI-powered transcription on my own terms.

If you value your time, your data, and your productivity—build this once, and thank yourself every week.

👋 About Me

Hi, I’m Shuvangkar Das, a power systems researcher with a Ph.D. in Electrical Engineering from Clarkson University. I work at the intersection of power electronics, DER, IBR, and AI — building greener, smarter, and more stable grids. Currently, I’m a Research Engineer at EPRI (though everything I share here reflects my personal experience, not my employer’s views).

Over the years, I’ve worked on real-world projects involving large scale EMT simulation and firmware development for grid-forming and grid following inverter and reinforcement learning (RL). I also publish technical content and share hands-on insights with the goal of making complex ideas accessible to engineers and researchers.

📺 Subscribe to my YouTube channel, where I share tutorials, code walk-throughs, and research productivity tips.

Connect with me: