witblack/whisper repository overview

⁠Docker of OpenAI Whisper (Voice to text (TTS))

This repository provides ready-to-use Docker images for OpenAI Whisper optimized for CPU-only environments. Each image is prebuilt with a specific Whisper model, eliminating runtime downloads and ensuring predictable performance, faster startup, and reproducible deployments.

The images are designed for batch transcription, offline usage, and server-side automation where GPU is not available or not required.

⁠Available Images & Tags

Each tag represents a fixed Whisper model baked into the image:

• witblack/whisper:large-v3
• witblack/whisper:medium
• witblack/whisper:small
• witblack/whisper:base
• witblack/whisper:tiny
• witblack/whisper:without-downloaded-model

⁠Tag Description

tiny / base / small / medium / large-v3 Image contains the Whisper runtime plus the model already downloaded.

without-downloaded-model Minimal image. No model included. Suitable for custom or dynamic model handling.

When to Use Which Image

tiny / base → Fast, low RAM, basic accuracy

small / medium → Balanced accuracy and performance

large-v3 → Highest accuracy, heavy CPU and RAM usage

without-downloaded-model → CI pipelines, custom workflows, or volume-mounted models

⁠How It Works

Whisper is installed at build time.

For model-specific tags, the model is downloaded during image build.

No apt update cache or temporary files remain.

Images are production-ready and immutable.

⁠Basic Usage

Transcribe an Audio File

docker run --rm \
  -v $(pwd):/data \
  witblack/whisper:small \
  whisper /data/audio.wav --language fa

Output files will be generated in the current directory.

⁠Specify Output Format:

docker run --rm
-v $(pwd):/data
witblack/whisper:base
whisper /data/audio.wav --language en --output_format srt

⁠Using the Image Without a Preloaded Model:

docker run --rm \
  -v $(pwd):/data \
  witblack/whisper:without-downloaded-model \
  whisper /data/audio.wav --model tiny

Model will be downloaded at runtime.

⁠Notes & Best Practices

CPU-only by design. No CUDA or GPU dependencies.

Larger models require significantly more RAM.

Recommended to mount data via volume (-v) instead of copying files into the container.

Best suited for backend services, batch jobs, and offline transcription systems.

⁠Summary

This repository provides clean, deterministic, and ready-to-run Whisper Docker images across all major model sizes, enabling fast adoption without manual setup or runtime surprises.