Add support in TextToSpeech for an audio stream type param and utterance

ID specified as a hashmap param in the synthesis calls.
Fix a bug where the cached parameters were not passed to the service
when synthesizing to a file.