pulaoecho-voice-assistant

The gateway streams PCM audio from multiple devices over TLS-encrypted WebSocket connections, providing per-device isolation for concurrent wake-word detection, ASR processing, and TTS response generation across all connected remote hardware.

Yes, you can use Sherpa-ONNX or Aliyun as selectable ASR engines for multi-device speech recognition. Both engines process incoming WebSocket audio streams and route asynchronous recognition results through OpenClaw integration for per-device response handling.

Per-device isolation in a WebSocket-based voice assistant architecture ensures independent audio stream processing, wake-word detection, and signaling workflows for each connected device. It maintains separate concurrent sessions so recognition and TTS playback routing never cross between devices.

Wake-word detection triggers per-device signaling workflows consisting of wake, process, feedback, and sleep states. As PCM audio streams through the WebSocket gateway, the system isolates the active session, runs ASR, generates TTS, and routes playback back to the specific originating device.

TLS encryption is required for WebSocket audio streaming in this multi-device voice assistant to secure PCM audio transmission. It protects wake-word detection, ASR, and TTS signaling workflows between remote devices and the centralized gateway during concurrent sessions.

WebSocket streaming for real-time TTS and ASR across multiple devices requires continuous PCM audio transmission and stable network connectivity. Latency depends on the selected ASR engine, and concurrent session scaling is bounded by gateway resources managing per-device isolation and signaling workflows.