Pipecat Realtime Voice POC

Idle

Transcript

Provider

Selects which realtime speech-to-speech provider backs the session.

Model

Selects the provider model for the next session.

Reasoning effort

Controls GPT Realtime 2 reasoning depth. Higher values may improve reasoning but can add latency.

Voice

Voice ID sent to OpenAI for generated speech.

Transcription model

Model used to transcribe caller audio for transcripts and context.

Output speed

Speech playback speed multiplier for OpenAI audio output.

Max output tokens

Maximum tokens OpenAI can generate in one response.

Shorter, more direct instructions usually make assistant speech easier to interrupt cleanly.

Turn eagerness

Controls how readily OpenAI decides the caller has finished a turn. Higher values respond sooner and can be more sensitive to brief pauses.

Stop assistant speech on caller audio

When enabled, OpenAI cancels the active assistant response as soon as caller speech is detected.

Cancel tool calls on interruption

When enabled, in-flight appointment tools are cancelled when the caller interrupts. When disabled, the tool finishes and the result can arrive later.

Tool timeout seconds

Maximum time the provider waits for a synchronous appointment tool result before timing out.

Provider-

Model-

Detail-

Audio-

Not loaded

Open this tab to load Roark results.