websocket

Send audio to a WebSocket server as binary messages. The sink accepts either raw audio (raw_audio) or an audio_encoder output (encoded_audio). It does not accept video, captions, or JSON.

For raw input, each message is interleaved signed 16-bit little-endian PCM (pcm_s16le). For encoded input, each message is one AAC or Opus access unit. The sink opens one WebSocket connection per input stream.

Example

{
  "name": "audio_ws",
  "type": "websocket",
  "inputs": ["pcm16k"],
  "config": {
    "url": "wss://example.com/audio",
    "headers": {
      "Authorization": "Bearer <token>"
    },
    "subprotocols": ["audio"]
  }
}

Use audio_resample before this sink when the server requires a specific sample rate or channel count.

To send encoded Opus instead:

{
  "nodes": [{
    "name": "opus",
    "type": "audio_encoder",
    "inputs": ["audio_mix"],
    "config": {"codec": "opus", "bitrateBps": 64000}
  }],
  "sinks": [{
    "name": "audio_ws",
    "type": "websocket",
    "inputs": ["opus"],
    "config": {"url": "wss://example.com/audio"}
  }]
}

Config

Field	Type	Default	Description
url	string	required	Absolute ws:// or wss:// URL.
headers	object	{}	Additional HTTP headers sent during the opening handshake.
subprotocols	string[]	[]	Ordered WebSocket subprotocols offered to the server.

The managed connection layer owns the outbound connection and reconnect policy. URL, headers, and subprotocols are not exposed to media processing components.

Handshake metadata

AVFlow adds these headers to each track connection:

Header	Description
X-AVFlow-Audio-Format	pcm_s16le, aac, or opus.
X-AVFlow-Audio-Codec-Config	Base64 codec initialization bytes for encoded input, when available.
X-AVFlow-Audio-Sample-Rate	Sample rate in Hz.
X-AVFlow-Audio-Channels	Interleaved channel count.
X-AVFlow-Identity	Participant identity, when available.
X-AVFlow-Source-Name	AVFlow source name, when available.
X-AVFlow-Track-Sid	Track SID or stable runtime track ID.
X-AVFlow-Track-Name	Provider track name, when available.
X-AVFlow-Track-Source	Track source such as microphone, when available.

Each WebSocket binary message contains only one PCM chunk or encoded access unit. AVFlow reconnects failed connections according to the job reconnect policy. During sustained backpressure it drops the oldest queued audio to preserve real-time delivery.

Pricing

Raw PCM has no audio encoding-duration fee. An audio_encoder input incurs the normal audio encoding-duration fee. In both modes, successfully written payload bytes are billed at $0.12/GB. Sink pricing.