Advanced Settings
Configure speech recognition, turn-taking, speech speed, background sounds, and input filtering in an Advanced Settings resource. Link it to one or more Assistants with advancedSettingsId.
Overview
Each configuration object is optional. Omitted settings use the platform defaults documented on the individual settings pages.
An Advanced Settings resource groups five configuration objects:
- STT controls transcript accuracy: keyword biasing and silence-based endpointing.
- VAD controls turn-taking: when the system considers the user to be speaking, when it considers them done, and whether the user can interrupt the Assistant.
- TTS controls how the Assistant speaks: the speech speed.
- Sounds controls background audio: the sound played while the Assistant is working on a response.
- Audio controls input filtering: the noise cancellation applied to the caller's voice.
Assistants linked to the same resource share its configuration. To use a different configuration, change the Assistant's advancedSettingsId.
Supported settings depend on the Assistant's architecture and providers:
- On an LLM + TTS + STT pipeline, STT and TTS support depends on the selected providers.
- With an STS (Speech-to-Speech) model, the realtime provider normally detects speech and turns.
stt.endpointinghas no effect.vad.endpointingMaxDelayandvad.minSpeechDurationhave no provider equivalent when provider-side turn detection is used. Support for other fields varies by model. - On GPT Realtime and Gemini Live, disabling interruptions switches turn detection to local VAD, making local speech detection settings effective.
- An STS Assistant with a separate TTS model uses that TTS integration's supported speed setting.
Unsupported fields are ignored. A configuration can save successfully for both pipeline and STS Assistants while producing different behaviour. See STT Settings and VAD Settings for provider support.
Data model
| Field | Type | Required | Description |
|---|---|---|---|
id | uuid | Auto-generated | Unique identifier |
name | string | Yes | Descriptive label for the configuration |
description | string | No | Free-text explanation of the configuration's purpose |
stt | object | No | Speech-to-text settings. See STT Settings |
vad | object | No | Voice activity detection settings. See VAD Settings |
tts | object | No | Text-to-speech settings. See TTS Settings |
sounds | object | No | Background audio settings. See Sounds Settings |
audio | object | No | Input audio filtering. See Audio Settings |
Creating an Advanced Settings resource
To create Project-scoped settings, use a project-scoped API key or include projectId with an organization-scoped key. With an organization-scoped key, omit projectId to create Organization-scoped settings. Data API paths below are relative to the server URL in the API reference, which includes /data_next.
Code
Code
Code
Attaching to an Assistant
Link an Advanced Settings resource to an existing Assistant by setting advancedSettingsId:
Code
Code
To detach, set the field to null:
Code
Updating Advanced Settings
Update the resource with PATCH. The change applies to every linked Assistant on its next conversation.
Code
Code
When changes take effect
In-progress conversations continue using the settings that were active when they started.