@langwatch/scenario
    Preparing search index...

    Type Alias SpeakerRole

    SpeakerRole: "user" | "agent"

    Voice recording, timeline events, and latency metrics — type contracts only.

    PR1 ships the type surface; PR2+ adds the runtime that populates these (writing segments + events on each adapter turn, computing latencies, saving WAV/MP3 via ffmpeg). Python parity: python/scenario/voice/recording.py.

    These are the output-side types attached to a ScenarioResult for voice runs: result.audio (a VoiceRecording), result.timeline (a VoiceEvent[]), and result.latency (a LatencyMetrics).