How redhat-et/whisper-self-hosted-llm works
This repository packages a self-hosted Whisper speech-recognition demo. A user uploads a WAV file in the Streamlit client, which posts it to the inference endpoint and displays the returned text. The README also describes Whisper’s multilingual recognition, translation, and language-identification capabilities and the available model sizes. The endpoint implementation and model-loading code are not among the sampled sources, so those internal details are not expanded here.
Main entry points: whisper-model-service Whisper inference service · whisper_client.py Streamlit client · models Whisper models
Components 4
User interface
- Streamlit client
whisper-model-service/streamlit/whisper_client.py
Inference
- Whisper inference service
whisper-model-service
Model artifacts
- Whisper models
models
Other components
- Audio uploader
Connections 4
- Audio uploader → Streamlit client: uploads WAV
- Streamlit client → Whisper inference service: posts audio
- Whisper inference service → Streamlit client: returns text
- Whisper inference service → Whisper models: uses model