Practical overview
When it fits
Use it when you want WhisperX in a Docker-based workflow and can prepare the documented GPU-enabled Docker environment. Its prebuilt tags suit users choosing from packaged model-language combinations; the no-model image is documented for selecting a model at runtime and sharing an alignment cache.
Who it is for
- Developers who want to run WhisperX transcription workloads in Docker.
- Teams packaging speech-recognition workflows around prebuilt model-and-language images.
- Users who need to build WhisperX images or manage a Docker-based transcription workflow.
Documented uses
- Run an audio file through WhisperX and request an SRT transcript.
- Choose a prebuilt image pairing a model and language, or select a model at runtime with the no-model image.
- Retain alignment-model downloads between containers by mounting /.cache.
- Build and maintain Docker images for WhisperX, including builds that incorporate the repository’s submodules.
Requirements & limits
The project provides prebuilt images through GitHub Container Registry and Dockerfile source through GitHub. The repository’s Dockerfile and CI workflow files are MIT-licensed; the underlying WhisperX program has a separate BSD-4 license.
- GPU setup is part of the documented run workflow, with NVIDIA-specific prerequisites varying by operating system.
- Image tags determine which model and language pairing is packaged; the no-model tag allows runtime model selection.
- The repository says its WhisperX code tracks a submodule commit and that a scheduled workflow targets the upstream main branch.
- For Taigi audio with the Breeze-ASR-26 image, the README says to skip alignment with --no_align because the bundled alignment model is trained for Mandarin phonology.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run an audio file through a prebuilt WhisperX image
Generate an SRT transcript from an audio file using a prebuilt model-and-language image.
Before you start
- Docker with GPU support configured and an NVIDIA GPU setup.
- On Windows: Docker Desktop, CUDA Toolkit, NVIDIA Windows Driver and WSL2. On Linux or macOS: NVIDIA GPU Driver and NVIDIA Container Toolkit.
- An audio file available in the directory mounted as /app.
Steps
Choose a published image tag matching the required model and language.
Run the container with GPU access, mount the current directory at /app, and pass WhisperX arguments after --.
Use --output_format srt and provide the audio filename, following the documented command pattern.
Expected outcome
WhisperX processes the audio in the container with SRT selected as the output format.
Guide limitations
- This workflow requires GPU-enabled Docker as shown in the project instructions.
- Available model-language combinations depend on the published image tags.
Run the no-model image with a selected model and shared cache
Select a WhisperX model at runtime while mounting a persistent cache location for alignment models.
Before you start
- Docker with GPU support configured and an NVIDIA GPU setup.
- An audio file available in the directory mounted as /app.
Steps
Use the no_model image tag, also referred to in the README as latest.
Mount the current directory at /app and a named volume at /.cache.
Pass the desired --model, --language and output-format arguments after --, followed by the audio filename.
Expected outcome
The container runs WhisperX with the selected model and language, while the mounted cache can be shared between containers.
Guide limitations
- The documented cache-sharing example uses the no-model tag.
- The guide does not specify a cache lifecycle or storage requirements.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- jim60105/docker-whisperX · Official homepage title · retained in its source language