I created my 3 samples of my wake-word in 16bits 16khz mono wav and when I go to validation by the instruction :
arecord -r 16000 -f S16_LE -c 1 -t raw | bin/rhasspy-wake-raven --keyword <WAV_DIR>
it gets stuck on this message :
arecord -r 16000 -f S16_LE -c 1 -t raw | bin/rhasspy-wake-raven --keyword /home/pi/rhasspy/rhasspy-wake-raven/Gaston
moqart
August 3, 2020, 10:58am
2
I think it is not stuck. It is just listening and trying to detect the wakeword from your samples. It will output something every time it detects your wakeword.
1 Like
Hi @arpagor62970 , add a --debug flag to your command-line to see more output printed. Otherwise, as @moqart said, it will only output a line of JSON when detection occurs.
I usually have to change to --probability-threshold to 0.45 (default is 0.5).
ok i did not understand that he was waiting for me to say the wakeword … i will try again
1 Like
Now I have to submit another problem: in the doc it is indicated that the wakeword wav files must be 16 bits and 16 Khz … so I used Audacity and created my 3 wav in 16 Bits -16 Khz
Now that I have my wav files which are well recognized by raven with the --keyword option I have modified my profile.json and replaced
“wake”: {
as indicated in the documentation by:
“wake”: {
and now I restart Rhasspy and I get this error:
DEBUG:rhasspyprofile.profile:Loading /home/pi/.config/rhasspy/profiles/fr/profile.jsonmain .py”, line 4, in main .py”, line 56, in mainmain .py”, line 80, in maininit .py”, line 97, in profile_to_confinit .py”, line 783, in print_wakeinit .py”, line 762, in get_wakemain
Looks like you have an older version of Rhasspy. Make sure you’re at 2.5.5.
the changes are so fast that I am always one step behind
1 Like
Playing with Raven in Rhasspy 2.5.5 hassio-addon! See strange behavior of Rhasspy:
No “Wake WAV” sound after the Wake up word recognition.
After termination of the session - sound of “Wake WAV” and sound of “Error WAV” plays twice.
Also no “Wake WAV” sound if I push “Wake UP” button in Web UI.
I installed the last version: 2.5.1 : rhasspy_2.5.1_armhf.deb
/.pycache pycache /fixup_keywords.cpython-37.pycpycache /publish.cpython-37.pyc
it’s normal ?
but the problem of the validation sound after the wake word as well as that emitted in the event of not understanding is identical: they are only emitted randomly but nevertheless the wake word is well detected and the intention well interpreted
Hi,
how raven works and does it get better after each success wakeword?
How can I record more samples?
It’s tuned to a specific person, so it would be best to record a second wake word for your son. It can be the same word, but his voice is probably going to be different enough.
Recording multiple wake words in the web UI is still a bit awkward, but it can done.
Thanks.
does raven get better after each success wakeword?
moqart
October 17, 2020, 5:02pm
15
No it does not get better with each recognized wakeword. It can be configured to save a recording of each activation which could be used to train another wakeword system like mycrof precise.
How good is raven working for you? Do you get a lot of false activation ?
I have no tried raven as i set my wakeword up before raven got introduced as an option.
