Rosanswers logo


I am looking for speech recognition packages. Someone can help-me with suggestions.

This is complex problem, it is required that applications are all the time listening and processing the information. This is not the core of my work but i need something that translates voice to text.

I have found (http://www.pirobot.org/blog/0022/ and https://code.google.com/p/ros-pocketsphinx-speech-recognition-tutorial/wiki/Tutorial) which uses pocketsphinx, but from what I have seen is not accurate and is required to build a dictionary of words.

Another solution is using the google framework (http://blog.achuwilson.in/2012/06/speech-recognition-in-ros-linux.html), but it is required internet connection and can take more the 3 seconds to get an answer.

Another option is use Julian (http://achuwilson.wordpress.com/2012/01/11/speech-recogition-using-julius-in-linux/) from video looks great.

The Hark seems to be powerful but they are using multiples microphones for localization also. (http://www.ros.org/wiki/hark, http://winnie.kuis.kyoto-u.ac.jp/HARK/wiki.cgi?page=HARK-ROS+Installation+Instructions)

I need something light then HARK and more efficient then pocketsphinx. I think that I will start with HARK. Can you give-me others suggestions?


Originally posted by Filipe Santos on ROS Answers with karma: 346 on 2013-04-09

Post score: 4

Original comments

Comment by Chik on 2013-04-10:
interesting question

Comment by Jordi Adell on 2013-04-18:
From my point of view PocketSphinx is a powerfull tool for Speech Recognition. However, it need some tuning to get the most of it. Also, HARK is a good think, I do not like the way the implemented but are good algorithms for localisation and separation in quiet environments.


1 Answer 1


Rosanswers logo

The most important here, is that you do not need to choose between HARK and Sphinx. HARK is a signal processing toollkit for source localisation and separation and Sphinx is a speech recognition software. However, it is very very likely that to combine them you will need to retrain or adapt the acoustic models.

Originally posted by Jordi Adell with karma: 26 on 2013-04-18

This answer was ACCEPTED on the original site

Post score: 1


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.