Admin Settings → Audio configures the listening (speech-to-text) and speaking (text-to-speech) capabilities users experience in their audio settings.
Speech-to-Text (dictation and voice mode)
| Option | Notes |
|---|---|
| Speech-to-Text Engine | Choose the transcription engine — a local Whisper-based engine (default, works offline) or an external provider. |
| Engine model | For local engines, select the model size; larger models are more accurate and slower. Leave empty to use the default. |
| Provider connection | External providers (for example hosted speech APIs) need the endpoint/key the engine requires. |
| Language locales | Restrict the languages users may dictate in. Auto-detect remains available otherwise. |
| Max Speakers | For multi-speaker transcription (meeting-style), the maximum number of speakers to separate. |
| Supported MIME Types | Which audio file types can be uploaded for transcription. |
| Output format / Response splitting | How transcriptions are returned; splitting affects long recordings. |
Testing: after changing engines, run a dictation test in chat; update the STT model if you changed its selection.
Text-to-Speech (read-aloud and voice mode)
| Option | Notes |
|---|---|
| TTS engine/provider | The speech synthesis provider configured for your deployment (local or hosted vendors such as ElevenLabs or Azure AI Speech, depending on your install). |
| Endpoint URL / keys / region | Provider connection details; keys are hidden and can be shown with the eye control. |
| Voices | The voice list users see. Availability depends on the provider connection being valid. |
| Default voice and settings | Starting values new users inherit. |
Note:
External speech providers send audio content outside your environment. In restricted deployments, use only local engines.
Common issues
| Symptom | Check |
|---|---|
| "No voices available" for users | The TTS provider isn't connected or its key is invalid. |
| Poor transcription accuracy | Use a larger local model, or set the explicit language locale. |
| Voice mode missing | Voice features are licensed/enabled per deployment — see Licensing. |