Error in C++ SDK

#21

@rjames
Thanks for reply. Maybe I understand your advise. The easiest way to achieve my goal is to install Ubuntu on my target,then install our edge-impulse-linux. The i can run the model.eim.
Cause i want to learn something from our Edge Impulse,i will try the traditional way to achieve it. Firstly, i need to install a compiler (e.g gcc-7.5.0) . Secondly, install the alsa lib development package (to get the headers) for my target. I need to cross-compiler them to my Aarch64 target without Ubuntu? Is there any instruction for me to refer.

#22

@y1165949971

So far you’ve tried edge impulse for linux C++ SDK but couldn’t get it to build because of not having the dependencies (alsa lib headers, C/C++ compilers) on your AARCH64 Linux target.

If you’re not able to get the C++ SDK running, there’s other SDKs you can try given you have the right dependencies:

  • If you have python on your target you can try the Linux Python SDK, follow instructions here and/or see this video.
  • Or if you have Node.js, you can use the Linux Node.js SDK,
  • Or if you have Go then use the Linux Go SDK.

Each provide you with the edge-impulse-linux-runner utility that can run your modelfile.eim.

#23

@rjames Hi rjames
I cross-compiler the ALSA-LIB on Ubuntu,and then link the libasound.so to the path.

image
image762×560 55.7 KB

Then i successfully build “AUDIO” file by run "APP_AUDIO=1 CC=aarch64-linux-gnu-gcc CXX=aarch64-linux-gnu-g++ TARGET_LINUX_AARCH64=1 USE_FULL_TFLITE=1 make -j
". Does this run successfully like this ? But maybe my microphone has some problem,it doesn’t change anyway.
image

#24

@y1165949971
Awesome work!

It can be either (1) unexpected behaviour of the model or (2) an issue with the audio/microphone.

To rule out (1)

  1. Build the APP_CUSTOM.
  2. Grab features from Studio.
    2.1 Go to Live Classifaction and choose a sample and Load sample
    2.2 Copy the Raw features and save it to a file called “features.txt”
  3. Run the custom app against the copied features and compare the results with studio.
    3.1 ./custom features.txt

Repeat the steps above with different features (copied from Studio). If it produces the same results then we can rule out (1) and the issue may be (2).

#25

@rjames Thanks for your sincerely help. You do help me a lot.
image

1 Like
#26

@y1165949971,

Great looks like the model is performing as expected. I’m glad you got it working

1 Like
#27

@rjames
I have solved the recording problem by setting .asoundrc file. And the model runs great!
I get another question about it. My project has three commands (forward, backward,stop). When I don’t say anything or I just say something else, it will also have a corresponding maximum probability. If I try to use it to control the progress of a car, it will cause the car to run around. This is the result when i speak nothing.
image
Of course, I can decide whether to control the car by setting the output probability threshold, but it does not solve the fundamental problem.
May I ask if it is possible to train an additional set of data as noise interference, and everything that has nothing to do with the three instructions I need is classified as noise.

Thanks~

#28

Hi @y1165949971,

Nice!! that you have the audio working.

For better visibility, for future posts can you please create a new topic on our forum? https://forum.edgeimpulse.com/

Regarding your question, yes you should add a new class i.e. noise and train your model to classify this class when it’s not any of the others. Take a look at our docs for tips on how to recognize sound.

#29

Yes i can . Thanks for your help