Voice recognition : What news?

By | August 23, 2015

Voice recognition had quickly evolve since few years and is now a sector in expansion. Everybody know voice recognition système ship with our phone. ( SIRI, Google Now, Cortana) that intensively use huge data centers. If the correct response rate to questions posed to those services is growing everyday, this is done by knock out the privacy of their users. Those solutions indeed needs a quantity more and more phénoménal of private data from their users to provide ever finer answers. The need to have permanent internet connexion is another limiting factor. The response time or the loss network can quickly become a problem for a regular or professional use of this tools.

I propose to you to make a quick tour of differents solutions that can work offline and save your privacy.

  1. S.A.R.A.H : behind this nice name, hides the project of a domestic assistant wrote in C# for the client part and JavaScript with Node Js for the server part. This is the project that seems to be the most mature. You can use lots of plugins to add functionality. It is not exempt of all reproaches with the dependence to Microsoft with Kinect and language choose to write it.( Note, you can use standard microphone but the results may be less accurate. For complex queries it still used google voice api service that does not saves your privacy.
  2. JASPER : This project is fully open source and wants to become your personal and domotic assistance fully customisable according to your needs. It let you define custom commands and associate them with actions to do. This can be simple vocales answers, web search, domotic contrôle or anything else. At origin,  JASPER has used Pocket Sphinx engine to work. Since, support of some engines were add like  Google, AT&T, Wit.ai and Julius. Note that only PocketSphinx and Julius engine works offline.
  3. CMUSphinx : Also Open Source, CMUSphinx is the project that develops PocketSphinx engine. With this vocal recognition engine, CMUSphinx propose also a Java Framework call Sphinx 4 that let you develop multiplatform autonome application. Language dictionaries are included for different languages but you will probably have to develop yours to get more relevant results.
  4. Api Windows : Like  S.A.R.A.H, you can use the windows API to build a off-line vocal recognition software. This API correctly manages lot of languages. The C# language is mandatory like the use of Windows system.

In conclusion, it’s now possible to create a powerful domotic system that uses vocale recognition. You will probably have to pass sometime to develop or customize the engine used to obtain correct answers. If the use of your personal data don’t bother you, the Google API offers great results.

see you soon.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.