openclaw - 💡(How to fix) Fix [Feature]: feat(volcengine): Doubao Realtime voice provider via gateway-relay transport

Summary

Add Volcengine Doubao Realtime as a talk.realtime.provider, using the gateway-relay transport over the Doubao Realtime Dialog API (WebSocket-based, end-to-end speech).

This is the Volcengine equivalent of the existing OpenAI realtime provider.

Problem

Users with Volcengine / Doubao API access cannot use native speech-to-speech Talk mode. The existing volcengine plugin supports text LLM and TTS, but has no realtime voice provider.

Proposed Solution

Implement a gateway-relay provider plugin under extensions/volcengine/:

Capabilities

capabilities: {
  transports: ["gateway-relay"],
  inputAudioFormats: ["pcm16-24khz"],
  outputAudioFormats: ["pcm16-24khz"],
  supportsBargeIn: true,
  supportsToolCalls: false,
  supportsSessionResumption: false,
}

Configuration

{
  "talk": {
    "realtime": {
      "provider": "volcengine",
      "providers": {
        "volcengine": {
          "appId": "<volcengine-app-id>",
          "accessKey": "<volcengine-access-key>",
          "model": "1.2.1.1",           // optional, defaults to "1.2.1.1"
          "speaker": "zh_female_vv_jupiter_bigtts"  // optional
        }
      }
    }
  }
}

Implementation

A ready-to-review branch is available: shushushv/feat/volcengine-realtime.

The implementation includes 12 unit tests for the binary protocol encode/decode logic and a fix for an edge-case bug in error frame decoding.

Notes

• The X-Api-App-Key header value is a protocol-level fixed constant per Volcengine official documentation — not a user secret.
• No WebRTC involved; this uses the existing gateway-relay transport path already established by the OpenAI realtime provider.
• The Doubao Realtime API endpoint: wss://openspeech.bytedance.com/api/v3/realtime/dialogue