libmsspeech

Library for Microsoft Cognitive Services speech recognition. For more details about usage, take a look at my blog post.

This is the very first version that works. Do not use it in any serious application yet!

Prerequisites

• libwebsockets at least v2.1-stable, v2.2-stable.
• json-c.
• libuuid

Building

autoreconf --force --install
./configure
make

Using

Start by running exampleProgram to learn how to use the library:

Usage: exampleProgram [OPTION...] <key> <language>
  -d            Produce debug output.
  -f FILE        Audio input file, stdin if omitted.
  -m MODE        Recognition mode:
  -p MODE        Set profanity handling mode {raw|masked|removed}. Default is masked.
            {interactive|dictation|conversation}. Default is interactive.
  -t            Request detailed recognition output.

To recognize a file:

exampleProgram -f <path to wav> -m interactive <your subscription key> en-us

On Linux, you can stream audio directly from microphone using Debian alsa-utils:

arecord -c 1 -r 16000 -f S16_LE | ./exampleProgram -m interactive <your subscription key> en-us

or perform long dictation on Steve Jobs Standford University commencement speech:

curl -L -s https://archive.org/download/SteveJobsSpeechAtStanfordUniversity/SteveJobsSpeech_64kb.mp3 | \
mpg123 -w - -m -r 16000 -e s16 - | \
./exampleProgram -m dictation <your subscription key> en-us

More explanation and details on how to use the library can be found in this blog post.