whisper.cpp

whisper.cpp

OpenAI's Whisper speech recognition models converted to ggml files for whisper.cpp, in sizes from tiny to large-v3-turbo, with smaller quantized copies.

Made by ggerganov on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds OpenAI's Whisper speech recognition models, converted to the ggml format so that whisper.cpp can load them. It is a set of model files, from tiny to large-v3-turbo. Several sizes also come in smaller q5 and q8 versions.

The task is automatic speech recognition, which turns spoken audio into text.

Who it suits: It suits people who run whisper.cpp and need the model files it reads. A small file fits a modest machine, and a large one trades disk space for a bigger model.

Industry | Who uses it | How they use it | Example
Software | Developers | Add speech to text to an app through whisper.cpp | Transcribing a recorded voice note
Media | Editors and podcasters | Turn recordings into text | Making a rough transcript of an interview
Accessibility | Tool builders | Give spoken content a written form | Captioning a short clip
Research | Speech researchers | Compare model sizes on their own audio | Testing tiny against small

Skip it if you use a hosted transcription service and never run models yourself.

Where to use it:

• whisper.cpp, the program these files are made for
• Windows, macOS or Linux, with Python for the download step
• Your own computer, so audio does not need to go to a hosted service

Use case | Who | What they get back
Transcribe audio | Anyone running whisper.cpp | Text from spoken audio
Pick a small file | People on modest hardware | A model of 31 to 75 MiB for the tiny size
Pick a large file | People who want a bigger model | Files of up to 2.9 GiB for the large sizes

How to set it up: You need a few things in place first.

• Python and pip, for the download command
• whisper.cpp itself, which reads these files
• Disk space for the sizes you pick

1. Install the Hugging Face hub package with pip.
2. Download the repository with the hf command.
3. Point whisper.cpp at the model file you want. The whisper.cpp project holds its run details.

pip install -U huggingface_hub
hf download ggerganov/whisper.cpp

How to use it: You choose one model file by size and by whether you want a q5 or q8 version, then hand it to whisper.cpp. The listing names tiny, base, small, medium, large-v1, large-v2, large-v3 and large-v3-turbo, plus quantized and .en versions of some.

Tool | What it does
tiny | The smallest model, 75 MiB on disk
base | A step up, 142 MiB on disk
small | 466 MiB on disk
medium | 1.5 GiB on disk
large-v3 | The biggest of the v3 files, 2.9 GiB on disk
large-v3-turbo | 1.5 GiB on disk, with q5_0 at 547 MiB and q8_0 at 834 MiB

Example: You are setting up speech to text on a small laptop and pick base-q5_1, listed at 57 MiB. You hand that file to whisper.cpp with an audio recording, and it produces the transcript as text.

For newcomers and experts: Start with the base file. It is small, so the download is quick, and you can judge the quality on one recording. Move up a size only if the text is not good enough.

The files are ggml conversions of Whisper, so they work with whisper.cpp and not with other Whisper loaders. The q5 and q8 files are smaller copies of the same models, for example 31 MiB against 75 MiB for tiny. The listing also includes a small.en-tdrz file.

Title: Disk size of some files in the listing
tiny: 75 MiB
tiny-q5_1: 31 MiB
base: 142 MiB
small: 466 MiB
large-v3-turbo-q5_0: 547 MiB
large-v3-turbo-q8_0: 834 MiB

Download the repository with hf
Pick a model file by size
Hand the file to whisper.cpp
Get the transcript as text

Good to know:

• License: MIT
• Cost to run: none named
• Data it reaches: it downloads model files from Hugging Face onto your computer

On Hugging Face, 13 November 2026

• Task: automatic speech recognition
• Downloads in the last 30 days: 0
• Likes: 1,697
• License: mit
• Fine-tunes: 6
• Tags: automatic-speech-recognition

Links

• Model: https://huggingface.co/ggerganov/whisper.cpp
• Files: https://huggingface.co/ggerganov/whisper.cpp/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ggerganov/whisper.cpp

Then follow the usage in its model card. Nothing in this kit runs by itself.