It's currently using the legacy URLFetcher API, and needs to be updated to use the SimpleURLLoader API to work with the network service.
I also plan to reword the chunked upload API to make the consumer do more of the work, and avoid unbounded buffers in the network process (Since speech recognition is the only consumer of the API, this doesn't require any extra code code, either).
The main issue, other than the chunked upload, is that the speech recognizer code lives on the IO thread, so we'll need to get it a URLLoaderFactory that can be used there, or a URLLoaderFactoryGetter of some sort. Need to figure out what is / is not worth the effort of doing here here, as Mojo pipes aren't free.
Comment 1 by mmenke@chromium.org
, Mar 30 2018