Sign inSign up

ragedunicorn/whisper

By ragedunicorn

•Updated 5 days ago

Public repository for ragedunicorn docker whisper images

Image
Machine learning & AI
0

881

ragedunicorn/whisper repository overview

⁠Whisper Docker Image

Docker Whisper

Whisper speech recognition with whisper-ctranslate2⁠ (faster-whisper/CTranslate2) - transcribe and translate audio and video, on CPU or NVIDIA GPU.

⁠Quick Start

# Pull latest version (cpu variant)
docker pull ragedunicorn/whisper:latest

# Or the CUDA variant / a specific version (see the Tags tab for the available versions)
docker pull ragedunicorn/whisper:latest-cuda
docker pull ragedunicorn/whisper:<whisper-version>-cpu-<build>

# Transcribe a file (model is cached in a named volume)
docker run --rm -v $(pwd):/tmp/workdir -v whisper-models:/home/whisper/.cache/huggingface \
  ragedunicorn/whisper:latest video.mp4 --model medium --compute_type int8 --output_dir .

On Windows, use the PowerShell form - ${PWD} with braces and the colon inside the quotes (without braces PowerShell parses $PWD: as a scope-qualified variable, as in $env:PATH, and rejects the argument), and a backtick instead of a backslash to continue the line:

# Transcribe a file (model is cached in a named volume)
docker run --rm -v "${PWD}:/tmp/workdir" -v whisper-models:/home/whisper/.cache/huggingface `
  ragedunicorn/whisper:latest video.mp4 --model medium --compute_type int8 --output_dir .

⁠Features

  • 🚀 Fast: faster-whisper/CTranslate2 transcribes up to 4x faster than openai/whisper at the same accuracy
  • ⚡ Two variants: portable CPU image (latest) and GPU-accelerated CUDA 12 image (latest-cuda)
  • 🎥 Video files work directly: mp4/mkv audio is decoded by PyAV - no separate ffmpeg step
  • 📦 Model cache: models download once into a mountable ~/.cache/huggingface volume
  • 🔒 Non-root: Runs as the unprivileged whisper user by default
  • 🏗️ Multi-platform: cpu variant supports linux/amd64 and linux/arm64 (cuda is amd64-only)

⁠Usage Examples

⁠Transcribe a video file
docker run --rm -v $(pwd):/tmp/workdir -v whisper-models:/home/whisper/.cache/huggingface \
  ragedunicorn/whisper:latest video.mp4 --model medium --compute_type int8 --output_dir .

PowerShell:

docker run --rm -v "${PWD}:/tmp/workdir" -v whisper-models:/home/whisper/.cache/huggingface `
  ragedunicorn/whisper:latest video.mp4 --model medium --compute_type int8 --output_dir .
⁠Generate SRT subtitles
docker run --rm -v $(pwd):/tmp/workdir -v whisper-models:/home/whisper/.cache/huggingface \
  ragedunicorn/whisper:latest video.mp4 --model small --output_format srt --output_dir .

PowerShell:

docker run --rm -v "${PWD}:/tmp/workdir" -v whisper-models:/home/whisper/.cache/huggingface `
  ragedunicorn/whisper:latest video.mp4 --model small --output_format srt --output_dir .
⁠Translate any language to English
docker run --rm -v $(pwd):/tmp/workdir -v whisper-models:/home/whisper/.cache/huggingface \
  ragedunicorn/whisper:latest interview.wav --task translate --model small --output_dir .

PowerShell:

docker run --rm -v "${PWD}:/tmp/workdir" -v whisper-models:/home/whisper/.cache/huggingface `
  ragedunicorn/whisper:latest interview.wav --task translate --model small --output_dir .
⁠GPU transcription (NVIDIA, CUDA 12 driver required)
docker run --rm --gpus all -v $(pwd):/tmp/workdir -v whisper-models:/home/whisper/.cache/huggingface \
  ragedunicorn/whisper:latest-cuda video.mp4 --model large-v3 --device cuda --output_dir .

PowerShell:

docker run --rm --gpus all -v "${PWD}:/tmp/workdir" -v whisper-models:/home/whisper/.cache/huggingface `
  ragedunicorn/whisper:latest-cuda video.mp4 --model large-v3 --device cuda --output_dir .

Not included: live microphone transcription (needs an audio device) and speaker diarization (needs PyTorch).

⁠Tags

This image uses semantic versioning with a variant infix:

Format: {whisper_ctranslate2_version}-{cpu|cuda}-{build_number}

⁠Version Examples
  • latest - Most recent stable release (cpu variant)
  • latest-cuda - Most recent stable release (cuda variant)
  • X.Y.Z-cpu-1 - whisper-ctranslate2 X.Y.Z, CPU variant, build 1
  • X.Y.Z-cuda-1 - whisper-ctranslate2 X.Y.Z, CUDA variant, build 1
  • X.Y.Z-cpu-2 - Rebuild of the same version (base image bump, fixes)

X.Y.Z is the whisper-ctranslate2 version. The available versions are listed in the Tags tab⁠.

When updates are available through automated dependency management, new releases are created with appropriate version tags.

⁠Links

⁠License

MIT License - See GitHub repository⁠ for details.

Tag summary

Content type

Image

Digest

sha256:0b9571a46…

Size

182.3 MB

Last updated

5 days ago

docker pull ragedunicorn/whisper