Exciting prospect for using your work...
I'm getting the following message for different talks I've tried transcribing:
Could not get valid response from Google servers for segment starting at second ...
Followed by an error like this:
/Users/brett/audio-conversion/node_modules/gspeech-api/index.js:63
full += timedTranscript[i].text + ' ';
^
TypeError: Cannot read property 'text' of undefined
I haven't taken a close look at your code, but could you just skip over errors (or provide them as HTML response strings on some error object property), or perhaps return an array of successful strings with error objects interspersed) and allow one to iterate oneself to decide what to do with error portions?