Turn detection profiles

A profile is a pre-configured turn detection mode. Each profile sets the defaults for a use case: you choose one when connecting by including it in the endpoint path, and the server handles the rest.

Agent STT is in Preview.

Compare the profiles

Profile	Turn detection	Best for
adaptive	Adapts to speaker pace and hesitation	General conversational agents
agile	VAD-based silence detection	Speed-first use cases
smart	adaptive plus ML acoustic turn prediction	High-stakes conversations
external	Manual — you trigger turn end	Push-to-talk, custom VAD, LLM-driven

adaptive

Endpoint: /v2/agent/adaptive

Adapts to each speaker's pace over the course of a conversation. It adjusts the turn-end threshold based on speech rate and disfluencies such as hesitations and filler words, waiting longer for speakers who tend to pause mid-thought.

Best for: general conversational applications.

Languages: all supported languages. Disfluency detection is available for the supported languages; other languages fall back to speech-rate adaptation.

Trade-off: latency varies by speaker.

agile

Endpoint: /v2/agent/agile

Uses voice activity detection (VAD) to detect silence and finalize turns as quickly as possible. This is the lowest-latency profile.

Best for: use cases where response speed is the top priority and occasional mid-speech finalizations are acceptable.

Languages: all supported languages.

Trade-off: because it relies on silence, it may finalize a turn while the speaker is still mid-sentence, for example during a natural pause. This can result in additional downstream language model calls.

smart

Endpoint: /v2/agent/smart

Builds on adaptive with an additional ML model that analyzes acoustic cues to predict whether a speaker has genuinely finished their turn. This is the most conservative profile and the least likely to interrupt.

Best for: high-stakes conversations where cutting off the user is costly, such as finance, healthcare, and legal.

Languages: Arabic, Bengali, Chinese, Danish, Dutch, English, Finnish, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Marathi, Norwegian, Polish, Portuguese, Russian, Spanish, Turkish, Ukrainian, Vietnamese.

Trade-off: higher latency than adaptive.

external

Endpoint: /v2/agent/external

Turn detection is fully manual. The server accumulates audio and transcript until you send a ForceEndOfUtterance message, at which point it finalizes everything spoken up to that point and emits an AddSegment.

Best for: push-to-talk interfaces, custom VAD pipelines, or setups where a language model decides when to respond.

Languages: all supported languages.

Trade-off: you are responsible for all turn detection logic.

Early turn signals

Two profiles emit predictions before the turn actually ends, which you can use to start preparing a response and reduce perceived latency:

• EndOfTurnPrediction — emitted by adaptive and smart
• SmartTurnResult — emitted by smart only, and higher-confidence

Turn detection in streaming

Streaming transcription has its own turn detection, configured with a silence threshold rather than a profile. The two are separate mechanisms: see Turn detection for the streaming equivalent.