Scenario: I've build an offline voice recognition service which generally wake up phones screen by detecting these phrases ("marco", "knock knock", "ding dong", "hey stupid") . But i can see that vosk can't give right output result more than 1.5ft or 2ft of distance, but can detect "hello" phrase more than 3/4ft of distance really well. What's actually i need to do for increasing the accuracy of detection if i want to cover more than 2ft of distance?
nb:
I already tried with different models (vosk-model-small-en-us-0.15, vosk-model-en-us-0.22 ,vosk-model-en-us-0.22-lgraph) but these models can't detect my above phrases well, like when i say "marco" they gives me output as "michael", "mark" etc etc.
I was figured out the accuracy issue by using grammer like below,
Hello dev! Here is the full config i'm using for android,
Version: 0.3.32 Model : vosk-model-small-en-us-zamia-0.5 Gradle: 7.3.3-bin.zip Android studio: Chipmunk (2021.2.1 Patch 1)
Scenario: I've build an offline voice recognition service which generally wake up phones screen by detecting these phrases ("marco", "knock knock", "ding dong", "hey stupid") . But i can see that vosk can't give right output result more than 1.5ft or 2ft of distance, but can detect "hello" phrase more than 3/4ft of distance really well. What's actually i need to do for increasing the accuracy of detection if i want to cover more than 2ft of distance?
nb:
I was figured out the accuracy issue by using grammer like below,
Recognizer(model, 16000.0f, "[\"marco\",\"knock knock\",\"ding dong\",\"hey stupid\"]")
but vosk giving me unexpected result like when i used to say in other language vosk detect it as marco, knock knock, ding dong, hey stupid.
Please help me guys and thanks to all for this awesome library.