Using IBM's STT

[Cqoicebordel]

I've tried a lot of STT, because my results are really poor. In one test, I used IBM's speech, and I wanted to share the results :
First, it's at https://www.ibm.com/watson/services/speech-to-text/. It's written nowhere, but you have 100min of transcribe free each 30 days. If the service is not used during 30 days, it's deactivated.
I won't describe how to signup, or to get the auth couple (it's not your account couple, you have to generate some specifically for the service).

Anyway, to test, I modified a few things I'll share here. I won't to a pull request, because I took shortcuts in the code that shouldn't be integrated to Jarvis, but they are few, so someone can correct what I did to integrate the code.
First, I modified /utils/configure.sh to add IBM in it, l64 :

command_stt)           options=('bing' 'wit' 'snowboy' 'pocketsphinx' 'ibm')

Then, I created the folder /stt_engines/ibm/, and wrote main.sh in it :

#!/bin/bash
_ibm_transcribe () {
    json=`curl -s -X POST -u xxxxxx-xxxxx-xxxx-xxxxx:PASSWORD --header "Content-Type: audio/l16; rate=16000" --data-binary @$audiofile "https://stream.watsonplatform.net/speech-to-text/api/v1/recognize?model=fr-FR_BroadbandModel"`
    $verbose && jv_debug "DEBUG: $json"
    echo $json | jq '.["results"][]["alternatives"][]["transcript"]' > $forder
}

ibm_STT () { # STT () {} Listen & transcribes audio file then writes corresponding text in $forder
    LISTEN $audiofile || return $?
    _ibm_transcribe &
   jv_spinner $!
}

As you can see there it's quite easy. Of course you have to change the couple username:password, and probably use a config file for it. I also used the model fr-FR_BroadbandModel hard written in the code. Adapting to the language would be nice. The doc is here : https://console.bluemix.net/docs/services/speech-to-text/input.html#models Finally, I used jq to parse the JSON, because I learned to use it for https://www.openjarvis.com/plugins/pluie-%C3%A0-une-heure. Using perl may be better as it means less dependencies.

For the results, it worked fine. Not better than Bing, and a tad worse than Wit. So, I won't use it. But it might suits better another language or another case of use. That's why I'm sharing it, so it may help maybe someone else.

[Oliv4945]

Hi,

Thank you for the example. What is missing to do a PR, is it time or skills ?

[Cqoicebordel]

Mostly time and laziness. I don't really want to spend some more time on it, knowing I won't use it. And knowing that anyone with knowledge of the project codes might integrate it in a tenth of the time it would take me. The todo-list isn't long and difficult, at least for someone familiar with the code of Jarvis :

• How to add config variables for login and password, and add the 'UI' to configure them
• Change jq to perl
• Handle the language from config
• Be sure there isn't other things to do to to integrate a new STT engine

[Oliv4945]

Ok, I was hopping that it was skills so I can help to to finalize integration :) There is nobody really knowing code as @alexylem is taking a little break, but if somebody want to do it I think that it can be added thanks to:

How to add config variables for login and password, and add the 'UI' to configure them

• Adding ibm_watson_user and ibm_watson_user to utils/configure.sh ->variables (L4)
• Adding two new cases (ibm_watson_user and ibm_watson_user) to the switch L55, with something like
  • eval "$1=\"$(dialog_input "IBM Watson username\nHow to get one: XXX" "${!1}" true)\"";;
  • eval "$1=\"$(dialog_input "IBM Watson password\nHow to get one: XXX" "${!1}" true)\"";;
• Adding a new if L273

      if [ $trigger_stt = 'ibm_watson' ] || [ $command_stt = 'ibm_watson' ]; then
          configure "ibm_watson_user"
          configure "ibm_watson_password"
      fi

• Adding IBM Watson settings to utils/menu.sh->options L140, then create a new case with

                                  IBM*)
                                      while true; do
                                          options=("IBM Watson username ($ibm_watson_user)"
                                                   "IBM Watson password ($ibm_watson_password)"
                                          case "`dialog_menu 'Settings > Voice recognition > IBM Watson' options[@]`" in
                                              IBM*Watson*username*)   configure "ibm_watson_user";;
                                              IBM*Watson*password*)   configure "ibm_watson_password";;
                                              *) break;;
                                          esac
  done;;

Change jq to perl jq is already installed and used by Jarvis "core", so it is not a big deal

Handle the language from config Can be obtained from ${language//_/-}, so the command should be

json=`curl -s -X POST -u xxxxxx-xxxxx-xxxx-xxxxx:PASSWORD --header "Content-Type: audio/l16; rate=16000" --data-binary @$audiofile "https://stream.watsonplatform.net/speech-to-text/api/v1/recognize?model=${language//_/-}_BroadbandModel"`

A variable could be added for BroadbandModel and NarrowbandModel

Ok so I did almost all the work, I might integrate it to Jarvis one day :)

[Cqoicebordel]

Except for the fact you wrote twice ibm_watson_user instead of ibm_watson_user and ibm_watson_password, it seems good to me :)

For jq vs. perl, it's because I looked at Google's STT. But you know better than me :)