I’m sure I’m doing something really dumb but those are the only two things he recognizes in offilne recognition. That is with the dictionary and keyword files that came with respeaker and I’m also trying to use the full CMU dictionary. Any thoughts or ideas?
Thanks
Jeff
Hi Jeff
Please refer to <LINK_TEXT text=“http://wiki.seeedstudio.com/ReSpeaker_C … ffline-stt”>http://wiki.seeedstudio.com/ReSpeaker_Core_v2.0/#play-with-offline-stt</LINK_TEXT> for the offline STT. thanks.
best rgds
Bill
Respeaker Core v2.0
Well yes, I did read it, over and over again. I also started over from scratch, rebuilding the image on the SD card and following each installation step carefully. What I discovered (so far) is that when I run microphone.py, it works better than the example program. I found that words in keywords.txt are recognized, mostly. Her is a sample of the keywords file I’m playing with:
respeaker /1e-30/
roboto /1e-30/ << does not recognize (what I’d like to change the wake up word to)
hello /1e-10/
goodbye /1e-30/
spaghetti /1e-30/
task /1e-30/
And here is some sample output:
Wake up
01111+100000000-00000000000000000000000000000000000000001111+1111111111111111111111111111111100000000-000000000000000000000000000000000000000000000000000Recognized task spaghetti
Wake up
00000000000000000000000000001111+1111111111111111111111111100000000-00000000000000000000000000000000000000000000000000000000000000000000Recognized task hello
Wake up
00000000000000000000000000000000000000001111+11111111111100000000-00000000000001111+11111111111111100000000-000000000000000001111+00000000-00000000000000000Recognized task hello
Wake up
00000000000000000000000000000000000001111+1111111111111111000111111111111111111111111111111111111111100000000-00000000000000000000000000000000000000000000000000000000000Recognized spaghetti hello task
Respeaker is actually reversing the order of the spoken words. That is; I said spaghetti task, hello task, hello task, task hello spaghetti.
Also, when I add the word ‘say’ to the list, respeaker does not respond at all.
I hope that’s clear enough. Let me know if you need anything else.
Thanks!
After doing some step-by-step debug tracing, I’m pretty sure respeaker never gets to the code that looks up words in the dictionary.
So, what am I doing wrong? Anybody? Ideas?
Hi there,
I am simulating the issue and will keep you posted on the result tomorrow. thanks.
Bill
Hi Jeff,
I simulated the issue and the wrong wake up rate is quite high. The pocketsphinx is quite old tech. We decided to remove the part from wiki. By the way, we are working with snips team on STT as well. for more info, please refer to https://snips.gitbook.io/documentation/installing-snips. Sorry for the inconvenience. thanks.
Bill
Thanks for your help. I’ll try Snips.
Hey, so I looked up snips and they are no longer allowing signups on their console.
I went through a lot of work reading through their documentation and then installing their snips scripts only to find that the console is closed.
Public access to the Snips Console is closed.
Should you need continued access in support of your ongoing commercial projects, please send an email to [email protected].
What do I do now?