After my previous smart home project with Snips.ai failed due to Sonos buying Snips, I was forced to find an alternative to offline speech recognition with Snips.ai.
Unfortunately, there were no real open source alternatives. At least I couldn't find one. So I took matters into my own hands. Project Sinos was born. ("Sinos" is a pure artificial word, derived from the Latin Sinus.)
Project Sinos
CMU Sphinx - How To Listen Offline
The first and apparently biggest problem was finding an open source software for offline speech recognition. To my astonishment, this was solved very quickly after I discovered CMU Sphinx. Developed by Carnegie Mellon University, it is the offline speech recognition tool. In addition, quite a few languages are supported (including German). It has little, very little competition.
After the first quick successes, however, there were considerable problems in recognizing specific words in order to identify intents. I was able to fix this quickly by creating my own language models.
Raspi, Talk To Me
After my Raspberry Pi 3 could listen again, he only had to be able to answer me. Again, I quickly found a solution in the form of Espeak. In the aftermath, however, this turned out to be not very promising. It is not without reason that the project asks for help in creating languages for Espeak.
Putting It All Together
Sinos is written in Python 3 and needs some preparation. I am currently working on a first installation script that will help the Raspberry Pi 3 (Buster) develop new skills.
What About The Sonos Use Case?
Oh yes, right - originally I wanted to control my Sonos sound system with voice commands via the Raspberry Pi 3 or the Rock Pi S. The current status very briefly: It works pretty well. Despite the fact that it's ironic that Sonos of all people bought Snips.ai while I was working with Snips on a Sonos controller. 😀
Kommentare
Kommentar veröffentlichen