search

MycroftAI / mycroft-precise

A lightweight, simple-to-use, RNN wake word listener
Apache License 2.0
842 stars 227 forks source link

Noise cancelling when listening #214

Open NewBermuda opened 3 years ago

NewBermuda commented 3 years ago

To increase the accuracy of wake word detection in noisy locations, I thought about applying noise cancelling to the raw audio before handing the audio to the RNN.

Do you think this is feasible for real time detection? Any suggestions on how to do it ? Could I use a python library to apply the noise reduction before giving it to the RNN in real time?

el-tocino commented 3 years ago

Generally, it's not worth it depending on the hardware you run on (pi3 for sure is too light), and you should depend on the STT system to do its job. https://github.com/Shb742/rnnoise_python can be used to handle rnnoise in python.

NewBermuda commented 3 years ago

Generally, it's not worth it depending on the hardware you run on (pi3 for sure is too light), and you should depend on the STT system to do its job. https://github.com/Shb742/rnnoise_python can be used to handle rnnoise in python.

I am using Jetson AGX Xavier, so from a GPU standpoint it might be able to work.