Raspberry Pi & IBM Watson Speech to Text

Make your Pi listen to your speech and transcribe the results to the console.

See the recently transcribed text occasionally change as Watson decides on a new translation based on added context.

Also see https://github.com/watson-developer-cloud/raspberry-pi-time-weather-demo for an example of using Text to Speech on a Raspberry Pi

Note: While this example should still work, it uses the older REST API, which is a bit more complicated than the newer Websocket-based API. The current recomendation is to use createRecognizeStream (example) instead of createSession + observeResult + recognizeLive. (This code should be updated sometime soon.)

Requirements:

• Raspberry Pi - Tested on a Model B 2 (1GB)
• Node.js - Tested and working on v0.12.7. (I'm using pre-compiled binaries from http://conoroneill.net/node-v01040-and-v0127-for-arm-v7-raspberry-pi-2-banana-pi-odroid-c1-available) It was tested on v4.0.0 but at least one dependency had issues compiling.
• Speech to Text service credentials - Log into bluemix.net, create a STT service instance, bind it to an app, click the "show credentials" link in the app's dashboard. Then put the username and password in either index.js or the STT_USERNAME and STT_PASSWORD environment variables. You can also put them into a .env file which will be automatically parsed by https://www.npmjs.com/package/dotenv
• USB Microphone - Tested with the following:
  • Kinobo X000NPSKOB
  • Logitech Mobile Speakerphone P710e (connected via USB, not tested over bluetooth)
  • PS3 Eye
  • StarTech Stereo Audio USB Sound Card ICUSBAUDIOMH
  • Should work with any other Microphone that plays nice with ALSA

Recommendations:

• FLAC - Compresses the audio before uploading - sudo apt-get update && sudo apt-get install flac -y
• LEDs - To indicate when the system is "working" and when it's recording. I'm using a Pimoroni Explorer HAT Pro, but a few LEDs and a breadboard would do. See lightshow.js for what colors go on what pins.

Setup

Download the code and run npm install in the directory.

Running

Run node index.js, wait for the red LED to begin blinking, and then start talking.

If you don't see any text after a moment or two, make sure that you're online, your mic is unmuted, and the arecord and aplay commands can record and play back sound.

Todo:

• Test on a PI 1
• Add lots of links and cleanup the documentation
• Shoot a video of it in action