Google doesn't offer a true method of continuous speech recognition so this app is not perfect, nor could it ever be, nor could any other app like it be perfect. Until Google updates their service API that is. I've found the web speech API to be inferior to the android API in terms of accuracy and usability despite it being a continuous recognizer.
The app is designed to restart the speech recognizer every time speech is fully transcribed or if the recognizer times out from lack of detected speech. This works pretty well however sometimes it cuts your first word off. If that happens you'll have to repeat what you said. If you repeat soon enough, the voice recognizer will have started at the proper time.
Check the video in this Play listing for an unedited and unrehearsed demo of the app running on my Samsung S3.
I have mixed feelings about the app. I like the idea of it with the kind of streaming captioning there is like you said some trouble with getting the first few words occasionally. I have auditory processing impairment and would like the ability to save the captioning for later use because of memory difficulties. Can you alter the software so people can save what's been captions? Thank you very much for the work you've done on this.