Interface BridgeMediaFrame

Frame payload as a base64-encoded string. Used by transports that can only carry text. Mutually exclusive with BridgeMediaFrame.Bytes; populate one.

Raw frame payload as binary. Preferred for in-process server-bridged drivers — no intermediate encoding. Mutually exclusive with BridgeMediaFrame.Base64; populate one.

The video's height in pixels, set and left out together with BridgeMediaFrame.Width.

Whether a decoder can start at this frame, with no frame before it. true for a still image (a JPEG or PNG decodes on its own); for a fragment of encoded video, whether its first video frame is a key frame. Absent where the producer does not know, and on a piece that carries no video frame (audio, an MP4 init segment); a consumer treats absent as unknown.

The payload's format, for example 'image/jpeg' for a video-in frame. Audio frames may omit it (they are in the session's declared input format); a video frame without one cannot be sent to a model that needs an image type, and the realtime driver drops it.

Which source a video-in or screen-in frame comes from, when the endpoint can have several (each participant's camera, a shared screen): an opaque key that stays the same for one source, for example 'participant:ada:camera'. A model that takes one video stream is shown one source at a time; this key is how the host tells sources apart and notices a switch. Absent on audio frames and where there is only one source.

A human-readable name for the BridgeMediaFrame.SourceID source ("Ada's camera", "Ada's screen"), for the note that tells the model what it is now looking at.

Optional per-speaker label for diarized inbound media (requires the provider's SpeakerDiarization feature). Lets turn-taking attribute speech to a named participant. Typically the participant's ExternalId or display name.

Optional epoch-millisecond capture timestamp for the frame, used for ordering, latency accounting, and silence-window calculations in the turn-taking policy. When omitted the consumer may stamp arrival time.

Which track this frame belongs to (kind + direction).

The video's width in pixels, where the producer knows it: an image's own width (after any rotation and scaling), or for a fragmented MP4 stream, the width its init segment gives the video track, on the init segment and on each fragment that carries video. Absent on audio and wherever the producer does not know it; a consumer treats absent as unknown.