QTrobot Vosk speech recognition (offline)

info

For offline voice recognition on QTrobot AI@Edge variant, please refer to QTrobot Riva speech recognition tutorial.

Overview

signal_cellular_alt  Level:  Advanced
track_changes  Goal:  learn how to use QTrobot Offline speech recognition
task_alt  Requirements:

•   Quick start with coding on QTrobot
•   Create a ROS python project
•   QTrobot interfaces using ROS Services

In this tutorial we will learn about how to use QTrobot Offline speech recognition.

Create a python project

First we create a python project for our tutorial. let's call it tutorial_qt_vosk and add the required python file:

cd ~/catkin_ws/src
catkin_create_pkg tutorial_qt_vosk rospy roscpp -D "Using Vosk speech recognition"
cd tutorial_qt_vosk/src
touch tutorial_qt_vosk_node.py
chmod +x tutorial_qt_vosk_node.py

QTrobot speech service

Following are some standard supported languages:

•  en_US (English)
•  fr_FR (French)
• de_DE (German)
• es_ES (Spanish)
• it_IT (Italian)
• pt_PT (Portuguese)
• nl_NL (Dutch)

Open the tutorial_qt_vosk_node.py file and add the following code:

#!/usr/bin/env python3
import sys
import rospy
from qt_robot_interface.srv import *
from qt_vosk_app.srv import *

if __name__ == '__main__':

    rospy.init_node('my_tutorial_node')
    rospy.loginfo("my_tutorial_node started!")

    # define a ros service
    speechSay = rospy.ServiceProxy('/qt_robot/speech/say', speech_say)
    recognize = rospy.ServiceProxy('/qt_robot/speech/recognize', speech_recognize)
    # block/wait for ros service
    rospy.wait_for_service('/qt_robot/speech/say')
    rospy.wait_for_service('/qt_robot/speech/recognize')

    try:
        # call a ros service with text message
        speechSay("Say something after the beep.")
        speechSay('#CAR HORN#')
        resp = recognize("en_US", ['blue', 'green', 'red'], 10)
        rospy.loginfo("I got: %s", resp.transcript)
        speechSay("You said %s " % resp.transcript)
    except KeyboardInterrupt:
        pass

    rospy.loginfo("finsihed!")

Tips

1. When you run the code or when you call the service from command line always wait a least 1 second before saying something, so that microphone has time to open.
2. It might not work on the first call when you switch the language, second one will be ok.
3. Using options for better detection of words. If you want to get exactly 'yes' or 'no', you just need to call the service with this options ['yes','no']. Even if you say some really long sentence, it will detect just 'yes' or 'no'.