Yes, as Jochen said, the current plan "is to strip out the parameters and send it to the default speech engine, for now", and this is the current implementation in the two pending CLs that Kirti refers to below. The "chrome://speech-recognition" prefix is used to ensure the namespace does not conflict with any other future implementations.
The spec refers to the serviceURI as a "generic URI" and describes various ways that the user-agent can implement it. I believe that implementing this behavior within the "chrome://speech-recognition" method/domain is conformant with the spec.
I also agree that a user-agent can implement a "full URL" (such as
https://example.com ) and remain conformant with the spec. However the spec does not define the protocol between the user-agent and server, so implementing this does not provide a useful feature today. (However, the spec contains an "Editor's Note that refers to potential future definition work, possibly using WebRTC, at which point a "full URL" may be useful.) Also, there may be security considerations as well.
In any case, Kirti's proposed change does not affect this (that is, Chrome doesn't currently implement a "full URL", but it could implement a "full URL" sometime in the future).