[anti-censorship-team] WebRTC Encoded Transform (or Insertable Streams) for media channels in Snowflake?

[David Fifield]

We have discussed how Snowflake uses WebRTC data channels (DTLS), while
some other WebRTC apps use media channels (DTLS-STRP), and how that
could be used in a classifier. See for example Section 2 of
https://censorbib.nymity.ch/#Fifield2016b. We have also entertained the
possibility of encoding Snowflake data into an audio/video stream, à la
FreeWave (https://censorbib.nymity.ch/#Houmansadr2013a) or CovertCast
(https://censorbib.nymity.ch/#McPherson2016a), in order to use media
channels in Snowflake.

I came across an API that may permit us to use media channel, without
the trouble of an audio/video codec. WebRTC Encoded Transform (formerly
Insertable Streams, I think) provides a way to get at and manipulate the
raw bytes of media frames.

https://w3c.github.io/webrtc-encoded-transform/
> This API defines an API surface for manipulating the bits on
> MediaStreamTracks being sent via an RTCPeerConnection.

The API defines RTCEncodedVideoFrame and RTCEncodedAudioFrame types that
give you an ArrayBuffer of the encoded media data:

https://w3c.github.io/webrtc-encoded-transform/#RTCEncodedVideoFrame-interface
	// New interfaces to define encoded video and audio frames. Will eventually
	// re-use or extend the equivalent defined in WebCodecs.
	[Exposed=(Window,DedicatedWorker)]
	interface RTCEncodedVideoFrame {
	    readonly attribute RTCEncodedVideoFrameType type;
	    readonly attribute unsigned long timestamp;
	    attribute ArrayBuffer data;
	    RTCEncodedVideoFrameMetadata getMetadata();
	};
https://w3c.github.io/webrtc-encoded-transform/#RTCEncodedAudioFrame-interface
	[Exposed=(Window,DedicatedWorker)]
	interface RTCEncodedAudioFrame {
	    readonly attribute unsigned long timestamp;
	    attribute ArrayBuffer data;
	    RTCEncodedAudioFrameMetadata getMetadata();
	};

One of the intended use cases of Encoded Transform is a end-to-end
encryption when then media streams pass through a middlebox that removes
and re-adds hop-by-hop DTLS-STRP layers:

https://webrtchacks.com/true-end-to-end-encryption-with-webrtc-insertable-streams/
https://jitsi.org/security/
> You can turn on end-to-end encryption (e2ee) as long as you are using
> Jitsi Meet on a browser with support for insertable streams.

The explainer document has a sample of inverting all the bits in an
encoded media frame:

https://github.com/w3c/webrtc-encoded-transform/blob/5c7ab84f4ce338f299172fa94af62ad3e2e7da74/explainer.md#code-examples
	// Receiver transform
	function createReceiverTransform() {
	    return new TransformStream({
		start() {},
		flush() {},
		async transform(encodedFrame, controller) {
		    // Reconstruct the original frame.
		    const view = new DataView(encodedFrame.data);

		    // Ignore the last 4 bytes
		    const newData = new ArrayBuffer(encodedFrame.data.byteLength - 4);
		    const newView = new DataView(newData);

		    // Negate all bits in the incoming frame, ignoring the
		    // last 4 bytes
		    for (let i = 0; i < encodedFrame.data.byteLength - 4; ++i)
			newView.setInt8(i, ~view.getInt8(i));

		    encodedFrame.data = newData;
		    controller.enqueue(encodedFrame);
		}
	    });
	}

I am not clear on the difference between Encoded Transform, Insertable
Streams, and WebCodecs, but here are some possible browser support
tables:

https://chromestatus.com/feature/6321945865879552
https://caniuse.com/?search=webcodec
https://developer.mozilla.org/en-US/docs/Web/API/WebCodecs_API