GoogleSpeechAPIConverter now works asynchronously for long-running inputs

[tyarkoni]

For audio clips > 1 minute, Google's Cloud Speech-to-Text API now requires users to use the asynchronous long_running_recognize method instead of the synchronous recognize method. This will require us to check the duration of the audio and make an asynchronous call if needed.

[tyarkoni]

@adelavega, is this a very recent change (doesn't look like it...), or have you been working around it in NeuroScout by chopping up audio files or doing something else?

[tyarkoni]

Ugh, this is really annoying from our perspective. It looks like the Speech-to-Text API doesn't even work with an external uri—it requires files to be in Google Cloud Storage. That's going to add a whole layer of complexity to pliers if there's no way around it...

[adelavega]

Well I haven't really been using it for one. I've been avoiding speech to text, as I can usually get a transcript. And when I did use it, I used IBM Watson.

[qmac]

I may open a PR soon for an really good and accurate speech-to-text API (😉) thats fairly easy to use (still asynchronous).