Practical overview
When it fits
Use it when a terminal-based Whisper workflow and CTranslate2 inference controls suit your audio transcription task.
Who it is for
- People who want to transcribe audio files from a command line.
- Users adapting workflows written for the original OpenAI Whisper CLI.
- Developers who need controls such as batching, quantization or loading a CTranslate2-format custom model.
Documented uses
- Generate a transcription from an audio file.
- Translate a Whisper transcription into English.
- Transcribe live microphone input.
- Apply VAD, diarization or inference configuration to a transcription workflow.
Requirements & limits
The project is published as a public GitHub repository under the MIT license. Its README documents installation through pip and a prebuilt Docker image.
- The CLI is designed to use the same command-line pattern as OpenAI Whisper.
- The translation task supports English as its target language.
- Batched inference uses VAD and ignores several standard Whisper parameters.
- GPU execution requires NVIDIA cuBLAS for CUDA 12 and cuDNN 9 for CUDA 12; CPU execution is also supported.
- A custom model loaded from a directory must be in CTranslate2 format.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install the package and transcribe an audio file
Use the documented Python package installation and CLI command to request a transcription.
Before you start
- Python package installation using pip
- An audio file and a Whisper model
Steps
Install the package with the command below.
pip install whisper-ctranslate2Run
whisper-ctranslate2 inaguracio2011.mp3 --model medium, substituting the path to your audio file.
Expected outcome
The CLI processes the audio file for transcription using the selected model.
Guide limitations
- The example does not specify output-format settings.
- GPU execution requires NVIDIA cuBLAS for CUDA 12 and cuDNN 9 for CUDA 12; CPU execution is also supported.
Run transcription with the prebuilt Docker image
Use the documented container workflow to process an audio file from a mounted host directory.
Before you start
- Docker
- An audio file in the current host directory
- GPU access is optional; the documented
--gpusargument can be removed when no GPU is available
Steps
Pull the image with the command below.
docker pull ghcr.io/softcatala/whisper-ctranslate2:latestRun the image with the current directory mounted at
/srv/files/, pass the audio file path inside the container, and set--output_dir /srv/files/as shown in the repository README.
Expected outcome
The container runs the CLI on the mounted audio file; the image includes the small, medium and large-v2 models.
Guide limitations
- The repository's example uses an NVIDIA GPU argument; remove it when running without a GPU.
- Models not included in the image require a derived image with the model preloaded or Docker volumes to persist and share model files.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- Softcatala/whisper-ctranslate2 · Official homepage title · retained in its source language