Check the transcription server
Start the vLLM example first and confirm that its audio upload returns a text field. Open WebUI must be able to reach this endpoint from its backend. Loopback works when both processes share the same host network.
Set the speech-to-text provider
In Admin Settings → Audio, select an OpenAI-compatible speech-to-text provider. Enter the transcription server’s base URL, model ID and, for an authenticated server, its key. Leave the audio request format set to multipart.
Settings for a new same-host instance
AUDIO_STT_ENGINE=openai
AUDIO_STT_OPENAI_API_BASE_URL=http://127.0.0.1:8000/v1
AUDIO_STT_MODEL=oruk/orukqwen-asr-1.7b-v2
AUDIO_STT_OPENAI_API_REQUEST_FORMAT=multipartDictate a test phrase
- Open a chat, allow microphone access and record a short phrase. Check the resulting input text before sending it.
- If Open WebUI runs in a separate container, configure an authenticated private endpoint reachable from the container. Inside the container, localhost identifies the container itself.
- On an existing installation, saved admin settings take precedence over initial environment variables. Update the Audio settings and try the same recording again.