Thanks for the video! When you uploaded data from your computer, did you record that using the phone/edge impulse tool that you demonstrated earlier, or was it done another way (e.g. recording and cutting up manually with an editor). Also, while less convenient, would it be more accurate to record with the actual microphone you're going to use on the dev board?
All custom keywords were collected with phone, and then splitted in the Edge Impulse UI (with the automatic segmentation tool). "would it be more accurate to record with the actual microphone you're going to use on the dev board?" => Yep, but also much harder to get data from many people. The signal processing side normalizes most data already fortunately.
Whe I try to "re-balance" it gave me an error: Failed to perform train / test split. The solution was jus put this text in the box: perform split. It worked for me. Thank you
Thanks for the tutorial. I followed it and have sucessfully created a dataset of 3 words. Then trained and tested a model that classsify new data in live classification of Edge impulse very well. But when I deployed it to ESP32-EYE, and ran it via cmd terminal, it does not show those 3 lables and only shows yes, no, noise and unknown labels. Any advise will be appreciated.
Thanks, that was very helpful. I'll definitely look into the docs for inference using built C++ library. It would be very helpful if you could make a video for that too. Thanks again. 🙂
Thank you for the video and easy to use software! I am currently trying to get a .onnx file out of it and don't know how, do you have a tip for me? Thank you!
there should be an option where you can mark the area of waveform and rise time, fall time too! if the waveform goes outside the designated area the it can be detected..... same for rise time!
It's going to be hard as the Pico does not have an FPU, and the DSP instructions (for calculating the spectrograms) require floating point math at the moment. But you can build all kinds of other interesting models with Edge Impulse and the Pico, see e.g. ruclips.net/video/LmSrY_pZho0/видео.html
Excellent tutorial! Thank you so much. I have got started :)
Thanks for the video!
When you uploaded data from your computer, did you record that using the phone/edge impulse tool that you demonstrated earlier, or was it done another way (e.g. recording and cutting up manually with an editor).
Also, while less convenient, would it be more accurate to record with the actual microphone you're going to use on the dev board?
All custom keywords were collected with phone, and then splitted in the Edge Impulse UI (with the automatic segmentation tool). "would it be more accurate to record with the actual microphone you're going to use on the dev board?" => Yep, but also much harder to get data from many people. The signal processing side normalizes most data already fortunately.
@@janjongboom7561 Nice, thank!
Whe I try to "re-balance" it gave me an error: Failed to perform train / test split. The solution was jus put this text in the box: perform split. It worked for me. Thank you
Thanks for the tutorial. I followed it and have sucessfully created a dataset of 3 words. Then trained and tested a model that classsify new data in live classification of Edge impulse very well. But when I deployed it to ESP32-EYE, and ran it via cmd terminal, it does not show those 3 lables and only shows yes, no, noise and unknown labels. Any advise will be appreciated.
Thanks, that was very helpful. I'll definitely look into the docs for inference using built C++ library. It would be very helpful if you could make a video for that too. Thanks again. 🙂
How do you inference using the c++ library i need urgent help :(
Excellent video!!
Very interesting, there are some tutorial to how optimize for Nicla Voice?
How do I make this work on a NODEMCU ESP8266? Do I need to connect a microphone module or similar?
Thank you for the video and easy to use software! I am currently trying to get a .onnx file out of it and don't know how, do you have a tip for me? Thank you!
Awesome video
there should be an option where you can mark the area of waveform and rise time, fall time too! if the waveform goes outside the designated area the it can be detected..... same for rise time!
Can I use this impulse model in Python script running on RPi Zero, to detect hotword?
Can i get data "Hello World"? thank you
Can i use this model in raspberry pi pico?
It's going to be hard as the Pico does not have an FPU, and the DSP instructions (for calculating the spectrograms) require floating point math at the moment. But you can build all kinds of other interesting models with Edge Impulse and the Pico, see e.g. ruclips.net/video/LmSrY_pZho0/видео.html