sample_ptt_realtime.json

{
    "_description": "Sample: Push-to-talk (PTT) mode with real-time model. EXPERIMENTAL — PTT is not fully supported by the Voice Live platform. VAD cannot be fully disabled, which may cause interference. See _known_limitations.",
    "_rationale": "PTT mode is useful when the client explicitly controls speech boundaries (e.g., button press). EOU is disabled to prevent premature end-of-turn detection. Server VAD is used as background with extended silence timeout to minimize interference.",
    "_status": "Experimental — not validated for production use",
    "_known_limitations": [
        "turn_detection=None is not supported by VoiceLive; a VAD config is always required",
        "VAD and PTT can interfere, resulting in ~83% audio response rate (vs 100% for VAD mode)",
        "Silence timeout is set high (2000ms) to reduce premature cutoff, but some interference may still occur",
        "These settings have not been validated by Voice Live engineering — treat as best-effort"
    ],
    "model": "gpt-realtime",
    "voice": "en-US-Ava:DragonHDLatestNeural",
    "voice_type": "azure-standard",
    "sample_rate": 24000,
    "noise_reduction": "azure_deep_noise_suppression",
    "echo_cancellation": "server_echo_cancellation",
    "vad_type": "server_vad",
    "silence_duration_ms": 2000,
    "use_eou_detection": false,
    "push_to_talk": true,
    "enable_barge_in": false
}