Speakr is a personal, self-hosted web application designed for transcribing audio recordings
Python
Document that TRANSCRIPTION_MODELS_AVAILABLE applies only to the active connector
Murtaza Nasir1a59925dc8b8
diff --git a/docs/admin-guide/model-configuration.md b/docs/admin-guide/model-configuration.mdindex cd236193..3cd463c3 100644--- a/docs/admin-guide/model-configuration.md+++ b/docs/admin-guide/model-configuration.md@@ -414,6 +414,11 @@ If the override isn't in `TRANSCRIPTION_MODELS_AVAILABLE`, it's silently droppedThe override is propagated to the connector via the `model` field on `TranscriptionRequest`. Connectors that key on a model name (OpenAI Whisper / Transcribe, Mistral, VibeVoice) honour it directly. The `asr_endpoint` connector forwards the override as a `model=` query parameter; the [whisperx-asr-service](https://github.com/murtaza-nasir/whisperx-asr-service) fork uses it to switch the loaded Whisper model on demand, while the upstream `ahmetoner/whisper-asr-webservice` ignores unknown query parameters, so the override is safe in either case.+!!! warning "List only models compatible with the active connector"+ The dropdown changes the model name **within the currently active connector**; it does not switch between providers. Speakr selects exactly one connector at startup based on `TRANSCRIPTION_CONNECTOR`, `USE_ASR_ENDPOINT`, and `TRANSCRIPTION_BASE_URL`, and every transcription request is routed there. If the list contains a model name the active connector does not recognise (for example, putting `gpt-4o-transcribe` in the list while `USE_ASR_ENDPOINT=true` is set), requests for that model fail with a 500 from the upstream service.++ For a WhisperX backend, list Whisper variants (`large-v3`, `medium`, `distil-medium.en`, etc.). For OpenAI's API, list `whisper-1`, `gpt-4o-transcribe`, `gpt-4o-mini-transcribe`, `gpt-4o-transcribe-diarize`. Mixing providers in one dropdown is a planned feature for a future release.+## Configurable Embedding ModelSpeakr's Inquire mode (semantic search) uses [sentence-transformers](https://www.sbert.net/) to embed transcript chunks locally by default. The default model is `all-MiniLM-L6-v2` (384-dim vectors), which is fast, small, and sufficient for most use cases.