SpeechFoundry on the Raspberry Pi 3 Model B

Follow the instructions below to get Inferret’s SpeechFoundry speech recognition running on your Raspberry Pi 3. Direct any questions or feedback to our speechfoundry-embedded-help Google Group. We plan to release new languages and updated models from time to time so please check back here.
The entire installation should take about 15 minutes depending on the speed of your internet connection.
NOTE: Although all speech recognition is performed locally on your Raspberry Pi, an internet connection is still required for this free release version and audio data will be uploaded to your SpeechFoundry account where you can review, transcribe and manage the recordings.

# Make sure you have a USB microphone (or USB adaptor and analog microphone) plugged in to your Raspberry Pi 3 Model B.

# Turn on your Raspberry Pi 3 Model B and boot into the Desktop

# Using the web browser sign up for an account at: http://accounts.speechfoundry.com/accounts/register/

# Check your email inbox for the activation email and click "Confirm my account" and log in (if not logged in automatically)

# Click the "RASPBERRY PI" tab at the top left of the screen

# Click the "Download installer" button

# Read the "Software Licence Agreement" and if you agree with the Terms click "Agree".

# You will then be prompted to download a zip-file containing an installation script and your user credentials. 
# Save this file in /home/pi/ or somewhere you can easily find it again.

# Open a terminal window and type the following:

cd /home/pi/

# Assuming you saved the inferret_rpi_installer.zip file in /home/pi/Downloads/ type:

cd /home/pi/
unzip Downloads/inferret_rpi_installer.zip

# This will create a directory called inferret_rpi_installer in which there are the following three files called: 
  • README.txt
  • installer.sh
  • credentials.txt
# To begin the installation process please change to the install directory and run installer.sh: cd inferret_rpi_installer/ ./installer.sh # The default location for installation is: /home/pi/SpeechFoundry/. You can change this during the installation process. # If there were problems with the installation look at the installation log in /tmp/log.out. # If the installation completed without problems you can start the English speech recognition system with: cd /home/pi/SpeechFoundry/scripts/ ./startSofia.sh en_US # or to start Japanese recognition use: ./startSofia.sh ja_JP # NOTE: To see recognition output for Japanese you will need to have installed the Japanese fonts during the installation process and re-booted your Raspberry Pi.
*) Lots of error messages like the following when running the ./startSofia.sh script:

perl: warning: Setting locale failed.
perl: warning: Please check that your locale settings:
	LANGUAGE = (unset),
	LC_ALL = "en_US.UTF-8",
	LC_CTYPE = "UTF-8",
	LANG = "en_US"
    are supported and installed on your system.
perl: warning: Falling back to the standard locale ("C").

To fix use the following commands and strings appropriate to the errors you are seeing:

sudo locale-gen en_US en_US.UTF-8
sudo dpkg-reconfigure locales

*) Lots of "Please wait a minute..." notifications of high latency

Are you seeing a small square in the top-right corner of your screen? If you are your Pi is not set up properly. If the square is "rainbow" colour, your power supply is under-voltage. You're probably using a spare USB telephone charger you had lying around. Don't! Get a proper stable USB power supply that will supply the correct voltage. Our ASR uses a lot of CPU when it's trying to work out what you're saying. If you see a small "red" square the CPU is getting too hot. You did put those heatsinks on, didn't you? No? If you see a small "yellow" square it means that CPU throttling is taking place again due to things getting too hot apparently down to only 600Mhz! That's why you're seeing all those latency warnings about not speaking. Put the heatsinks on, direct a fan at your Pi if necessary and cool things down. The warnings should go away although you'll still see them from time to time if you talk too much.