Can you please make a tutorial on how to get custom wake words to work on web? I have the demo app working but cannot for the life of me figure out how to drop in a new ppn/rhv to add a custom word.
Is it possible to have a wildcard as a slot? The usecase would be something that would trigger a search, for example: "I'd like to look up WILDCARD:queryPhrase", I'd like to get all the words said after the word 'up' and have it assigned to the queryPhrase variable. Even better would be to have wildcard mixed in the command for example, "I'd like to look up WILDCARD:queryPhrase and email me the results".
Good question. It depends if the text is confined enough and known ahead of time. With Speech-to-Intent, instead of transcribing speech to text, the speech recognition and language understanding are fused. This means much better accuracy and the resource efficiency to run on a microcontroller (!). But it also means it cannot do open-domain transcription. You'd need to also use a STT engine like Cheetah (and a bit more resources, like a Raspberry Pi) to accomplish that: picovoice.ai/products/cheetah/
looks very promising. good job
Can you please make a tutorial on how to get custom wake words to work on web? I have the demo app working but cannot for the life of me figure out how to drop in a new ppn/rhv to add a custom word.
Hello, I have problem can u help me out..
How can i download or add .ppn file in my project
i already add .rhn file but i don't found .ppn file
Is it possible to have a wildcard as a slot? The usecase would be something that would trigger a search, for example: "I'd like to look up WILDCARD:queryPhrase", I'd like to get all the words said after the word 'up' and have it assigned to the queryPhrase variable. Even better would be to have wildcard mixed in the command for example, "I'd like to look up WILDCARD:queryPhrase and email me the results".
Good question. It depends if the text is confined enough and known ahead of time. With Speech-to-Intent, instead of transcribing speech to text, the speech recognition and language understanding are fused. This means much better accuracy and the resource efficiency to run on a microcontroller (!). But it also means it cannot do open-domain transcription. You'd need to also use a STT engine like Cheetah (and a bit more resources, like a Raspberry Pi) to accomplish that: picovoice.ai/products/cheetah/