← Directory

Faster-whisper

Faster-whisper is a reimplementation of OpenAI's Whisper speech-to-text model built on CTranslate2, a fast inference engine for Transformer models.

  • AI
  • Tools & Utilities
Faster-whisper is a reimplementation of OpenAI's Whisper speech-to-text model built on CTranslate2, a fast inference engine for Transformer models. This container runs faster-whisper as a Wyoming protocol server, delivering up to 4× faster local transcription with lower memory usage for voice assistant setups.

Platform availability

Available on 2 of 9 platforms

  • Unraid
  • TrueNAS (not listed)
  • Umbrel (not listed)
  • ZimaOS
  • Proxmox (not listed)
  • Helm (not listed)
  • Mac (not listed)
  • Windows (not listed)
  • Linux (not listed)

Health score?A 0-100 score computed nightly from four factors: maintained (40%: recent commits and steady activity), popular (25%: GitHub stars, log scale), easy to install (20%: platforms, install notes, setup guides, Docker image) and light to run (15%: minimum RAM, ARM support). When the source code has been checked, a fifth factor, security check (20%: automated scan for committed secrets, unsafe desktop-app settings, known-vulnerable dependencies and project hygiene; any serious finding caps it at 20), is added. Factors without data are left out and the rest are rescaled.

74/100 Good

Maintained
75last commit 6 days ago
Popular
8825,776 GitHub stars
Easy to install
452 install platforms, install notes, Docker image
Light to run
602 GB minimum RAM, runs on ARM
Security check
95no serious findings, CI, tests

Availability: platform listings confirmed today, links checked 9 days ago

GitHub stars
25,776
Open issues
33
Last commit
2026-10-06
Latest release
v1.2.1
License
MIT
Activity
rising

Checked 4 days ago - source: GitHub (SYSTRAN/faster-whisper)

Resources & Compatibility

Minimum requirements: 2 GB RAM, 4 CPU core(s). ARM (e.g. Raspberry Pi) is supported.

Checked 15 days ago - source: Docker Hub image tags and published docs

First-install notes

  • Install the stable package first Use pip (Python's package installer) to run pip install faster-whisper. That gives Python the transcriber and its required parts, so your program can actually start and load a model. source
  • Use GPU settings only if you have one If you have an NVIDIA GPU, start the model with device="cuda" and compute_type="float16" to make transcription faster. If you use that on a CPU-only machine, it can fail or run poorly. source
  • Point to your local model folder If you downloaded a faster-whisper model to a folder, point WhisperModel at that folder, for example "/path/to/converted/model". This makes the app use your local copy instead of fetching a model name from an online source. source

Fetched 15 days ago - each note links to its source

Alternatives

No alternatives collected yet.

Common questions

How do I install faster-whisper for use in a self-hosted script?

Install the stable faster-whisper package with pip. It provides the transcriber and required dependencies so your program can start and load a model.

Can I use a pre-downloaded faster-whisper model folder instead of fetching a model name?

Yes, point WhisperModel at the local converted model folder. This makes faster-whisper use your local copy instead of fetching a model name from an online source.

What model settings should I use on an NVIDIA GPU?

Start the model with device set to cuda and compute_type set to float16. Using those settings on a CPU-only machine can fail or run poorly.

How can I expose faster-whisper as an API server?

Use an OpenAI-compatible faster-whisper server implementation, which wraps faster-whisper as the transcription backend. This lets you call it through an OpenAI-style transcription API.

Answers sourced from deepwiki.com, deepwiki.com, www.reddit.com

Community

Collected 14 days ago - public community threads