ROS Offline Speech Recognition

Offline speech recognition based on Vosk

System requirements:

• Ubuntu 20.04
• ROS Noetic

For other systems, the ROS node needs to be run as Python3 node.

Installation

Install python requirements:

pip3 install -r requirements.txt

Download models

cd ros_offline_asr
./download_models.sh

Configuration

These are the following parameters that can be passed to ROS node:

models_dir:     /path/to/models # By default its ./models
microphone_id:  "default"       # Microphone identifier for sounddevice library
sample_rate:    ""              # Sample rate, by deefault it will use device default sample rate

For microphone selection see: sounddevice.query_devices

Model names should correspond to language codes in dynamic reconfigure parameters: cfg/OfflineAsr.cfg

Dynamic Reconfigure

Following dynamic reconfigure parameters supported:

enable:  false
langauge: en-US # see cfg/Offline.asr for all options

Launching the node

roslaunch ros_offline_asr offline_asr.launch

Topics published

ROS node publishes the following topics

offline_speech (std_msgs/String) - Publishes final result of speech recognition.

offline_interim_speech (std_msgs/String) - Publishes interrim results as they are here