Whisper ASR for daily dialogue with standardized OpenAI API speech interface. Whisper ASR is a general-purpose speech recognition toolkit. Whisper Models are trained on a large dataset of diverse audio and is also a multitask model that can perform multilingual speech recognition as well as speech translation and language identification.

docker-compose up -d

• CPU

docker run -d -p 9000:9000 \
  -e ASR_MODEL=base \
  -e ASR_ENGINE=openai_whisper \
  onerahmet/openai-whisper-asr-webservice:latest

• GPU

docker run -d --gpus all -p 9000:9000 \
  -e ASR_MODEL=base \
  -e ASR_ENGINE=openai_whisper \
  onerahmet/openai-whisper-asr-webservice:latest-gpu

To reduce container startup time by avoiding repeated downloads, you can persist the cache directory:

docker run -d -p 9000:9000 \
  -v $PWD/cache:/root/.cache/ \
  onerahmet/openai-whisper-asr-webservice:latest

• Download english speech media sample

wget -O test.map3 https://www.cambridgeenglish.org/images/153149-movers-sample-listening-test-vol2.mp3

• Test Asr Api by media sample

curl http://172.25.0.2:9000/audio/transcriptions \
  -H "Content-Type: multipart/form-data" \
  -F '[email protected]' \
  -F model="whisper-1"

• Multiple ASR engines support (OpenAI Whisper, Faster Whisper, WhisperX)
• Multiple output formats (text, JSON, VTT, SRT, TSV)
• Word-level timestamps support
• Voice activity detection (VAD) filtering
• Speaker diarization (with WhisperX)
• FFmpeg integration for broad audio/video format support
• GPU acceleration support
• Configurable model loading/unloading
• REST API with Swagger documentation

Key configuration options:

• ASR_ENGINE: Engine selection (openai_whisper, faster_whisper, whisperx)
• ASR_MODEL: Model selection (tiny, base, small, medium, large-v3, etc.)
• ASR_MODEL_PATH: Custom path to store/load models
• ASR_DEVICE: Device selection (cuda, cpu)
• MODEL_IDLE_TIMEOUT: Timeout for model unloading

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies
uv sync --upgrade

# Export ffmpeg
export PATH="${PATH}:${PWD}/bin"

# Run service
uv run webservice.py --host 0.0.0.0 --port 9000

After starting the service, visit http://localhost:9000 or http://0.0.0.0:9000 in your browser to access the Swagger UI documentation and try out the API endpoints.

• This software uses libraries from the FFmpeg project under the LGPLv2.1