Whisper Batch Transcriber

Automatically convert all of your voice recordings into clean, organized, neat text files.

It's fully automated, unlimited, using state-of-the-art speech-to-text technology. Works 100% offline on your computer, privately.

Convert speeches, podcasts, webinars, monologues, storytellings and other audio speech into a formatted .txt file. Here’s what a sample output looks like from a generated .mp3 file:

There are many different types of text that can be used just to fill space.
Some of this text is about how people are treated unfairly in the world.
People want to be free and have control over their own information, but it's not always easy.
There are laws and rules that are supposed to protect people's rights, but they're not always followed.
Ultimately, the goal is to be able to live our lives freely and make our own choices without being controlled or judged..

Usually it's one sentence per line, with a capitalized first letter and period at the end. Easy to read.

How to Use:

1. Put your audio files into the folder called:  📂 put your audio files here
2. Run the transcriber by double-clicking: ▶ whisper_transcriber.py
3. The transcriptions will be saved as .txt files inside: 📂 output_transcripts

That's all! Easy as pie.

You can choose between two Whisper AI models:

• Small_en - requires 2GB of VRAM, 96% accuracy
• Large_v3_Turbo - requires 6GB of VRAM, 99,8% accuracy

The results with both models are REALLY GOOD. They’re relatively similar in speed, so if you do have a powerful enough computer to run Large_v3_turbo, then always use that.

Both models are pre-included in the package, so you don’t have to download them separately.

Install Info:

Before using this program, it requires having Python 3.10 or newer installed, plus a couple of script dependencies. You'll be running the program directly from the whisper_transcriber.py script file.

I've included a fully automatic one-time installer .exe that'll do the setup for you so that you'll be able to simply double-click the .py to run the program directly from hence forth.

The reason this is necessary is because I discovered compiling this python program into an exe actually makes it work slower than it would be if it was ran directly as a script.

Thankfully its easy to set up the system and I've made it as hassle-free and guided as possible. You'll find the needed info in the included Manual inside the folder.

System Requirements:

This program was tested on Windows 7, it requires 2GB of storage space and atleast a mid-range gaming Windows PC with a GPU that has a minimum of 2GB of VRAM, preferably more. 

6GB or more GPU VRAM is necessary to run this program at its full potential.

NOTE: The program is slow to start. It can take about 5-15 minutes before it fully loads up; you'll first see a console window with two warnings and nothing else during this time but eventually the welcome screen will appear. This is normal.

Worth Knowing: 

Accepted formats: ".mp3", ".wav", ".flac", ".ogg", ".m4a", ".aac"

Many audio files = Takes longer.

Longer audio files = Takes longer.

Files other than .wav = Takes longer.

Depending on your computer, 1 minute of audio can take about 2-10 minutes to process.

It will also use your CPU/RAM/VRAM to some degree, so you should let it work and not play performance-heavy 3D games in the background or do something else that is computationally expensive, like video editing. These activities can interfere with the process.

Accuracy and quality should be excellent, near perfect even, but mistakes can still happen. Use the Large-v3-Turbo model for best accuracy results.

I’d recommend throwing only about 120 minutes worth of audio into the program at a time, let it complete and then do another batch. Make sure to remove the previously completed copies of the audio files from the audio folder. There’s a chance the process might sometimes fail, so it’s best to not throw 100's of files and 10's of hours of audio all at once. Then again I haven't tested it with that many files, so it might actually work fine.

Protip: The program will run faster if your audio files are already .wav format, since that's the format the program actually needs to transcribe the speech-to-text. All other formats get converted into a temporary .wav file during the process so you can skip that if your files have been pre converted to a mono WAV file with a 16,000 Hz sample rate and 16-bit audio depth. Fre:ac, the free program to convert audio, can let you do this fairly easily.

https://www.freac.org/

Protip2: If your computer goes into sleep mode while this program is busy transcribing your files, you can prevent that by getting DontSleep.exe from SoftwareOk, a wonderful German software company:

https://www.softwareok.com/?seite=Microsoft/DontSleep

Credits for Whisper goes to OpenAI, Claude for coding assistance.

Other Transcribers:

https://offlinetranscribe.de/ (This uses the same underlying Whisper tech as mine, but is faster due to using a C++ version of it and it is more robust, but costs 50€/year.)

https://elevenlabs.io/blog/meet-scribe (Eleven Labs’ quality is EVEN BETTER than Whisper, but also costs $5/month per 12 hours of speech to text audio conversion. As of 2025, it's probably the BEST Speech-to-Text there is.)

The benefit of my program is that it's 100% free, unlimited, has good quality results, but is slower and less pretty.

Support My Work:

If you enjoyed this release, please buy me an orange to fuel me: https://buymeacoffee.com/reactorcoregames

Or join my Patreon for games, assets, design knowledge and tool recommendations: https://www.patreon.com/ReactorcoreGames

Check my Itch.io page and Follow me there to know when I release cool new stuff! https://reactorcore.itch.io/

All my links - I make games, software, assets, lego mechs, AI art and more: http://www.reactorcoregames.com

Join my Discord server to discuss my projects and get sneak peeks: https://discord.gg/UdRavGhj47 

Another way to help me is to share my things with your friends, school, work, family or on social media. Every bit of visibility helps a lot! 

Enjoy! 

- Reactorcore

result = subprocess.check_output(['nvidia-smi', ...]) result.decode('utf-8').strip() 

import sys if sys.version_info < (3, 8) or sys.version_info >= (3, 13):     print("Warning: This script is tested on Python 3.8–3.12 only.") 

# requirements.txt openai-whisper>=20231117 pydub>=0.25.1 torch>=2.0.0 numba>=0.59.0