Voice recognition module for ROS2 foxy

General
,
1

Hi,
Are there any packages for voice recognition in ROS2 Foxy?
Preferably to work offline like the PocketSphinx.

Thanks,
Ajay

2

I’m not sure about something ROS2 specific, but there’s Mozilla Deep Speech, which should run offline:

3 Likes
3

Audio support in ROS is pretty thin on the ground, but there are a few packages that will bind gstreamer pipelines

This ROS1 package looks like it would be pretty easy to port to ROS2:
http://wiki.ros.org/pocketsphinx
Otherwise, a riff on gscam might be doable, the pocketsphinx pipeline is pretty simple.
gst-launch-1.0 autoaudiosrc ! audioconvert ! audioresample ! pocketsphinx ! fdsink fd=1

I’d love to plug my own ros-gstreamer package, but it doesn’t support string payloads yet.

1 Like
4

Hi @Ajaykumaar_S,

Not offline. However, at AWS we have build a number of ROS 2 packages to help integrate audio recognition and speech generation with AWS cloud services like Amazon Lex and Polly. You can find them here:

Regards,
Cam

4 Likes
5

I just got pocketsphinx running with ROS2 and discovered it doesn’t like my accent.
Fortunately pocketsphinx is not the only speech-to-text package that has gstreamer bindings

It’s easiest to launch from bash, but that repo also has tools to launch it from roslaunch.

gst-launch-1.0 --gst-plugin-path=install/gst_bridge/lib/gst_bridge/ autoaudiosrc ! audioconvert ! audioresample ! pocketsphinx ! queue ! rostextsink

6

Another option would be the voice assistant Jaco:

It can run completely offline on most Linux computers and even on a RaspberryPi and supports multiple languages. For a robot project of mine I created an interface skill that allows to control the assistant over the ROS2 topics. You can find it as demo skill here.

If you’re only interested in Speech-To-Text without NLU or TTS, the STT module Scribosermo can also be used as standalone model.

1 Like