Our take
Quietno commits in the last 90 days and no release in a year- Activity
- 19
- Community
- 40
- Issues
- 21
- Pull requests
- 20
+3 stars in the last 30 days
This is a focused Whisper CLI with useful inference controls, but its quiet maintenance record and highly concentrated contributions make it worth validating against your environment before relying on it for ongoing workflows.
Good fit if
- You want a command-line interface built on OpenAI Whisper and Hugging Face's inference stack.
- Your machine has an NVIDIA GPU or Apple Silicon Mac and you can tune batch size to fit available memory.
- You are comfortable with the Apache-2.0 license and the project's current maintenance pace.
Look elsewhere if
- You need active recent development or regular releases: there have been no commits in the last 90 days and no release in over a year.
- You need a broadly maintained project with contributions spread across its contributors; most of the code comes from one person.
- You need support for hardware beyond the NVIDIA GPU and Mac platforms described by the project.
All health signals
| Last commit | 2025-10-25 (12 months ago) |
|---|---|
| Commits, last 90 days | 0 |
| Releases, last 12 months | 0 |
| Contributors | 20 (top contributor: 84% of commits) |
| Issues closed, last 90 days | 0 |
| Pull requests merged, last 90 days | 0 |
| Project age | 3 years |
Checked on 2026-10-25 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
13,057Forks
953Change since Nov 17, 2026: 0 stars·-1 forks
Overview
insanely-fast-whisper is an opinionated terminal CLI for transcribing audio on-device with OpenAI Whisper. It is built on Hugging Face Transformers and Optimum, with optional Flash Attention 2, and targets NVIDIA GPUs and Apple Silicon Macs. The goal is to make high-throughput Whisper transcription straightforward from a command line, rather than requiring users to assemble an inference pipeline themselves.
You can pass an audio file path or URL, choose a Whisper checkpoint, and save a transcript. The CLI also exposes translation, timestamps, and speaker diarization options. See the repository README for installation details and the full list of arguments.
Key Features
- Transcribes audio from a local file path or URL using OpenAI Whisper checkpoints.
- Uses Hugging Face Transformers and Optimum, with optional Flash Attention 2.
- Supports CUDA devices and Apple Silicon through the MPS backend.
- Offers batching and model selection, including distil-whisper checkpoints.
- Can transcribe or translate audio and produce chunk- or word-level timestamps.
- Provides optional speaker diarization using a Pyannote model and a Hugging Face token.
- Includes a Python pipeline example for users who prefer not to use the CLI.
Use Cases
- Researchers or developers processing long recordings who want to run Whisper inference from a terminal on compatible hardware.
- Mac users who want a command-line transcription workflow using Apple Silicon's MPS backend.
- Teams generating transcripts with timestamps or speaker labels for recorded conversations, where the required model access and credentials are available.
- Developers comparing Whisper model and inference settings on their own supported hardware.
What you need
Detected in the repository
- Python >=3.8 (from pyproject.toml)
- A test suite
License in plain words
Apache-2.0permissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the license, the NOTICE file and a note of your changes
- Share your changes: no
- Includes an explicit patent grant from the contributors:
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
Install with pipx and run the CLI on an audio file or URL:
pipx install insanely-fast-whisper
insanely-fast-whisper --file-name <filename-or-URL>
On macOS, pass --device-id mps. See the README for installation caveats, options, and Flash Attention setup.
Alternatives
- whisper.cpp: whisper.cpp transcribes with a lightweight C/C++ Whisper implementation for offline apps across desktop, mobile, browser, and server, rather than a Transformers-based CLI.
- whisper-web: whisper-web runs Whisper in the browser with Transformers.js, instead of using a local terminal CLI geared toward NVIDIA GPUs and Apple Silicon.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| insanely-fast-whisper | Jupyter Notebook | Apache-2.0 | 13.1k | Quiet |
| whisper.cpp | C++ | MIT | 54.1k | Active |
| whisper-web | TypeScript | – | 3.3k | Inactive |
Considerations
- The README describes support for NVIDIA GPUs and Mac MPS, not a general CPU-first workflow. GPU memory limits can require reducing the batch size; the README notes that MPS can be more memory-hungry than CUDA.
- Flash Attention 2 requires separate installation steps and is optional. The README includes a specific pipx installation command and troubleshooting notes.
- The project is marked Quiet: it has had no commits in the last 90 days, no release in over a year, and no recent issue closures or merged pull requests in the supplied health data. Although it has multiple contributors, contributions are concentrated heavily in one person.
- The repository metadata reports tests, but no CI configuration. Check the current code and dependencies before adopting it in a production workflow.
- The project is licensed under Apache-2.0. Review the license terms and any applicable model or dependency terms for your use.
Found this useful?
Share it with someone who would like insanely-fast-whisper.