interview-pilot-media-domain

name	interview-pilot-media-domain
description	AI-Meeting media and realtime communication skill. Use for Mimo ASR, TTS, WebSocket push, heartbeat, error notifications, and realtime audio message protocols; trigger on `/api/ip/v1/mimo/**`, `/api/ip/v1/websocket/**`, legacy `/api/ip/v1/xunfei/**`, or realtime `@ServerEndpoint` work.

interview-pilot-media-domain

Read Order

1. references/api-map.md
2. references/realtime-asr.md
3. references/tts.md
4. references/websocket-notification.md
5. references/object-dictionary.md
6. references/gotchas.md

Key Entry Points

• admin/src/main/java/com/interviewpilot/media/api/WebSocketController.java
• admin/src/main/java/com/interviewpilot/media/api/XunfeiTtsController.java (legacy class name, Mimo-backed implementation)
• admin/src/main/java/com/interviewpilot/media/infrastructure/websocket/AudioTranscriptionWebSocketHandler.java
• admin/src/main/java/com/interviewpilot/media/infrastructure/integration/MimoAudioService.java
• admin/src/main/java/com/interviewpilot/media/infrastructure/integration/AudioTranscriptionService.java
• admin/src/main/java/com/interviewpilot/media/infrastructure/integration/XunfeiAudioService.java (legacy disabled by default)
• admin/src/main/java/com/interviewpilot/media/infrastructure/integration/XunfeiLongTextTtsService.java (legacy disabled by default)

Invariants

• WebSocket connection session and business sessionId are not the same concept.
• WebSocket transcription must authenticate before starting a transcription session.
• A single WebSocket session must not stack unbounded transcription contexts.
• Command/event fields must stay aligned between frontend and backend.
• Mimo TTS is synchronous. /tts/tasks and /tts/tasks/{taskId} keep a task-shaped compatibility surface, but new code should read audioBase64 from the synthesis response instead of polling for provider-side completion.
• New code should use /api/ip/v1/mimo/**; /api/ip/v1/xunfei/** is compatibility only.