New user here and I searched this board for information about recording audio samples related to building a dataset.
Goal for my first project is to recognize the word ’ help ’ when my old mother falls on the floor of her appartment and then trigger some wireless alarm.
I’m expecting to need multiple mic’s/devices listening in different rooms, maybe some kind of mesh set up.
If I use a recorder (like the tascam-dr40x) to record audio samples from her appartment (like background noise) and her saying the word ’ help ’ in different situations (loud / soft / far away / close / tv on / shower ect) and later use a different quality mic on a mcu board… will this work ?
Hi @BoRRoZ, welcome! (I think I remember your username from the TTN forums by the way ).
Yes, the normalization step that happens in the signal processing pipeline will normalize most of this (here’s some background), just make sure that you use a microphone on the MCU with proper gain, because loudness can make a difference (or do another normalization step beforehand).
Regarding “help” - this might not be an ideal word due to a high number of false positives. Something with more syllables (“BoRRoZ help”) would be more reliable.
OK so in an optimal situation you should record the trigger words with the same type of mic used on the MCU board ?
The problem I forsee is that the quality of these mics are bad in comparison to the devices used for recording.
Well… I don’t know if I can learn my old mother other trigger words then ‘help help’
anyway… I hope to learn and build a lot with these new techniques
Hi Pete , actually, I didn’t even started (except buying some stuff)
Busy finishing my new workshop and another project, and mainly, catching up reading.
I have many ideas using this new technology, So after the summer … yeah !
(I’m very curious about power consumption for example)
).
I’m stuck
