SurfaceTextureHelper currently calls its listener with a texture id, a transform matrix, and a timestamp. This is close to being a complete frame, but not quite. The listener has to call back into SurfaceTextureHelper, passing the texture resolution, in order to create a frame.
SurfaceTextureHelper has now been updated to output org.webrtc.VideoFrames directly, and the old methods have been deprecated. These methods will be removed in two weeks.