wrapper for Snowboy: a hotword detection engine

I guess you need to play with the values set for the sensivity
Code: Copy 
sensivity = sensivity&"0.5"
It sets a sensivity of 0.5 for all the models.
 
Multiverse app said:
Yup. Your APK is working fine.
Click to expand...
Try this APK:
https://drive.google.com/file/d/0B4g9tud5lvjgRmhNSHFHclhNVzQ/view?usp=sharing

It will/should recognise the following:
Code: Copy 
    'alexa.umdl                         say "alexa"                        1
    'snowboy.umdl                       say "snowboy"                      2
    'Extreme.umdl                       say "extreme"                      3
    'smart_mirror.umdl                  say "smart mirror"                 4
    'pilot.umdl                         say "pilot"                        5
    'OK_Google.umdl                     say "OK Google"                    6
    'hello.umdl                         say "hello"                        7
    'Pluto.umdl                         say "pluto"                        8
    'stitch.umdl                        say "stitch"                       9
    'chris.umdl                         say "chris"                        10
    'barrier.umdl                       say "barrier"                      11

The number to the right is what the toast message should show when you pronounce the word according to what it can recognise. Pilot and Hello seems to be close to one another. But is seems to recognise the others (including Extreme) without too much difficulty.
 

Attachments

Syd Wright said:
Good that so many hotwords have now been generated. However, I think that adding more hotwords will substantially increase the chance of false triggers, simply because Snowboy will have to respond to a multitude of voice sounds.
Probably it is best to let the user chose only one hotword and only include/activate the corresponding umdl file.
What do you think?
Click to expand...
I have added Alexia, Amalia, and Alexander - but I think it is starting to get confused with names that are so close to one another:
Code: Copy 
    'the following are available thus far:
    'alexa.umdl                         say "alexa"                        1
    'snowboy.umdl                       say "snowboy"                      2
    'Extreme.umdl                       say "extreme"                      3
    'smart_mirror.umdl                  say "smart mirror"                 4
    'pilot.umdl                         say "pilot"                        5
    'OK_Google.umdl                     say "OK Google"                    6
    'hello.umdl                         say "hello"                        7
    'Pluto.umdl                         say "pluto"                        8
    'stitch.umdl                        say "stitch"                       9
    'chris.umdl                         say "chris"                        10
    'barrier.umdl                       say "barrier"                      11
    'Hi.umdl                            say "hi"                           12
    'Alexia.umdl                        say "alexia"                       13
    'Amalia.umdl                        say "amalia"                       14
    'Alexander.umdl                     say "alexander"                    15

Here is the link to the APK file

https://drive.google.com/file/d/0B4g9tud5lvjgYnlaaVVpOXRSNTA/view?usp=sharing
 
Syd Wright said:
Any more ideas (or experiments) to improve the speech recognition results?
Click to expand...
for each user, allow them to record their voice on snowboy dashboard. After the process, the page will download a pdml file. programmatically copy it to the snowboy folder and rename the extention to udml. Then, Re-initialize the snowboy object. This way, the recognition becomes better for that particular user.
 
Multiverse app said:
allow them to record their voice on snowboy dashboard
Click to expand...
the problem I am facing is that it appears that I cannot record the voice on a phone. The process has to be done in a desktop browser.
Have to figure it out.
 

Attachments

  • Screenshot_20170903-190202.png
    Screenshot_20170903-190202.png
    117.8 KB · Views: 548
Just realized that snowboy has been dumping its recorded data in "recordaudio.pcm" under DirRootExternal. The problem is that the file builds up over time and takes a huge amount of space. It was taking 41 GB of space when I noticed it.
So, is it a better choice to delete the file on Sub Snowboy_stop() ?
 
the - additional - "logging" was a result of debugging/trying the library....
As i am asuming you are using my version; I´ll adapt the library and send you a new one.

PD: You should start creating a NEW thread for any new question in the future. This thread is not the right place. this is the Joboffer forum.
 
Syd Wright said:
I would also like to join in the development program and receive the latest version by Don Manfred by making a donation. Has Don thought about how to proceed with such additional developers?
Click to expand...
The trial (finished library) is already uploaded here... If it goes in the direction you need; just donate as much you want.
 
Last edited:
Syd Wright said:
The ability to allow the user to select only one particular hotword
Click to expand...
It is up to you to just initialize the hotwords you want. You also can list the files in the folder and let the user select which you want to use...
Only initialize the hotwords you want (see example). You are not forced to use all available. In the example i load them all.
 
So if I needed a wrapper for iOS as well as Android?, could you help on this DonManfred? :)
 
You must log in or register to reply here.
Back
Top