Read the statement by Michael Teeuw here.
MMM-voice
-
@dmwilsonkc if you run your mycroft core from the node helper as i do with pocketsphinx in voice then there is no need at all to mess with the socket server
-
@strawberry-3.141 I didn’t realize that was a possibility. I currently start Mycroft when the pi boots up in a virtualenv as was recommended by the mycroft folks. Both the MM and Mycroft run exactly as they should. The only exception is the microphone issue- not being able to access the microphone from two processes at the same time. I have no idea how I would start Mycroft from the node_helper.js.
At this point I am very confused on the best way to go. To me, it seems like the socket communication between two concurrently running processes would be easier. But you are the expert.
Any advice you suggest would be welcomed. Thanks again.
-
Hi everyone. I’m having trouble with my voice module.
It doesn’t seem to register any of my commands. The microphone icon is blinking but whenever I say “hide modules” or any other command it doesn’t register it.
In the terminal it says “Starting pocketsphinx.” Don’t know it that is of any concerns? If anybody has any tips they are more than welcome.
-
@ime90 can you check the .lm and .dic files in your module folder and see if they are empty or include your commands
-
@strawberry-3.141 hi strawberry, thanks for the response.
I’ve managed to make it work, but the problem is that the operations were uuuuultra slow.
I’ve timed the response time in 10 different occurrences.
After saying “MAGIC MIRROR” it takes an average of 7 seconds for the microphone icon to start blinking.
After saying “HIDE” it takes an average of 15 second (hope this isn’t normal) to hide the modules.
Is this due to my RPI3? Any and all tips and experiences are welcome.
-
@ime90 I am running MMM-voice on a RPi3b and while it is not as fast as my Mycroft-core in responding to its wake word, it is much faster than the times you have stated. I turned the debug = true so that I could see what it is doing. I would say the times you have stated are at least double what it takes MMM-voice on my Pi. I changed my wake word to Hey Jarvis. I don’t know if that made any difference in how fast it recognizes the wake word. I also added the Hello-Lucy modifications to the MMM-voice module. It is not “fast” in responding to commands, but it only takes a few seconds (5 to 6sec.) to Hide/Show modules if I remember right.
-
I wrote this in a different thread not too long ago actually, but I noticed a vast increase in detection/response rate when moving from a cheap USB mic to a decent webcam microphone. I would say the response now is on the order of 1-2 secs, with effectively no errors in voice interpretation. I’m sure this is helped immensely because the clarity of the audio into the rpi is much clearer - the cheap USB mic had terribly noisy audio in.
-
i am running MMM-voice on an ODroid Xu4, which is very fast.
I also use another voice enabled smart mirror on this same device.
with a Logitech webcam as the mic.
MMM-voice gets the hotword correct every time, but almost none of the following words.
the same hardware setup works correctly with arecord as the wav source, but uses google speech detection for everything after the hotword.
it appears the system is using the default language model (lm) and dictation (dic) files.
anyone have a suggestion on debugging this?
-
@sdetweil set
debug: truein the config and you will see what the mirror recognizes -
i have that, but the text ‘recognized’ is not what i said.
if I speak slowly, i can get ‘magic mirror’
then wake up is ‘wake go’ or ‘wake all’,hide calendar is ‘hide who’
open help is ‘open go’the smart-mirror project also uses ‘go to sleep’ and ‘wakeup’
‘hide newsfeed’ is ‘hide to show the’, ‘hide is the’
none of the commands, when recognized, do anything…
hide modules
go to sleep (not on PI so the tvservice doesn’t exist), but no error is logged
