@langwatch/scenario
    Preparing search index...

    Function audio

    • Inject a pre-recorded audio file (WAV/MP3/OGG/FLAC) or raw bytes as the user's next turn. Bypasses the user simulator and TTS entirely.

      Files are auto-converted to PCM16 @ 24kHz mono by shelling out to the bundled ffmpeg binary (see resolveFfmpegPath; Python parity with imageio-ffmpeg — no system ffmpeg required). Remote URL-like strings (http://, rtmp://, etc.) are rejected so ffmpeg never issues outbound network requests on the caller's behalf.

      Parameters

      • pathOrBytes: string | Uint8Array<ArrayBufferLike>

      Returns ScriptStep