Voicemails suck. They're expensive and dull. Also if you disable it, with some carriers there is no way to now if someone called you while your phone is off.
What it does
It seems like a normal voicemail, except that it sends you the transcript of the message so you don't have to listen to it.
How we built it
The core app is built in Java. It waits for a phone call on Nexmo, answers it like any voicemail does the feeds the audio to the python code that sends it to Microsoft and returns the text to the Java code. Finally, spoofing the caller ID, the transcript of the message is sent via SMS to the user with date and time.
Challenges we ran into
Microsoft Cognitive API has some quirks when used with python. The documentation is not up to date and very poor especially for python. The effect that some parameters have on the speech recognition was not clear. Not all the parameters were documented. Nexmo records an mp3 while Microsoft wants a wav.
Accomplishments that we're proud of
We started with no ideas and made something that is actually useful and works.
What we learned
We learned a lot about Microsoft Cognitive and Nexmo APIs.
What's next for X called Y(ou)
We want to finish this service because it actually solves a problem we have. It would be nice to test multiple SpeechToText APIs and find the best one, to reduce the amount errors in the recognition.