Using speech-to-text to fully check out during con calls
Switch branches/tags
Nothing to show
Clone or download
joshnewlan Merge pull request #2 from cclauss/patch-1
Create AUTH once and then reuse
Latest commit 1befb16 May 28, 2016
Failed to load latest commit information.
Muted.m4a Initial commit. May 17, 2016 number nicely in markdown too May 24, 2016 Update May 27, 2016 Update May 17, 2016

Say What?

This script listens to meetings I'm supposed to be paying attention to and pings me on hipchat when my name is mentioned.

It sends me a transcript of what was said in the minute before my name was mentioned and some time after.

It also plays an audio file out loud 15 seconds after my name was mentioned which is a recording of me saying, "Sorry, I didn't realize my mic was on mute there."

Uses IBM's Speech to Text Watson API for the audio-to-text.

Currently relies on Splunk as a data store, but can be extended to use an open-source tool instead.

Relies on Uberi's SpeechRecognition PyAudio and API wrapper:

##Installation (OS X)

  1. Sign up for, install, and run Splunk Enterprise
    • This has to be enterprise; the HTTP Event Collector feature used here doesn't exist in light
  2. Add your credentials to CREDS in
  3. Add an http event collector to Splunk and enable it in global settings (but do not enable acknowledgement)
  4. [Get a hipchat API token](https://[your company] and update the hipchat fields in
    • Your hipchat user id is the second number in your hipchat jabber info
  5. Update name in unless your name is Josh
  6. Create an IBM Bluemix account
  7. Add a speech-to-text plan
  8. Add your credentials to for IBM_USERNAME and IBM_PASSWORD
  9. Install Homebrew
  10. brew install python
  11. brew install portaudio
  12. pip install pyaudio
  13. pip install SpeechRecognition
  14. pip install requests


  1. Start Splunk
  2. Run

TODO: Improve usage