Personal Audio Classifier

Problem Description: making voice / sound recognition as described in "Personal Audio Classifier" but the final app does not work, just sits in the initial screen (ILL 2 below)

  1. Device: OnePlus 6T with Android 11
  2. Personal Audio Classifier extension version: 20200904 (downloaded from here), bcs other users found out that the original version (provided with the original post) was obsolete
  3. Audio/recognition model created on Personal Audio Classifier site: 80-90% accuracy
  4. Tested device (smartphone) by installing "Test_PAC.apk" and it was OK
  5. Tested the app by compiling it and installing the "apk" on my smartphone (didn't use the companion)
  6. Blocks are different from the original instructions bcs this is just a test to see if the extension performs. The scenario is that only I will speak into the mic and the model should recognize me and return my name and the confidence level as the first pair in the results list (if recognized, my name = the first item in the first list within the results list). Please see ILL 1

ILL 1 - the blocks

ILL 2 - smartphone screen-shot

Questions:
Q1. The original instructions don't show a step where the classification is called to start (initiated), in other words, is it by design that it is not necessary to initiate the classification in some way and solicit the approval from the user - to use the microphone (for recording) ?

Q2. The AppInventor offers "call PersonalAudioClassifier1.ClassifySoundData" block, but the explanation says that it performs its function "on the image"....
Is that a typo or did the author had in mind an "image file" a recording of an audio signal - saved at some location where it can be pulled from" ? If so, what shall the audio format of that recording be ?

ILL 3

Q3. Most likely something is not right with my blocks, and it would be much appreciated if someone can point me into right direction.

Thank you
Milan

Did you try your aia on another device. The PAC does not work with all Androids as stated in several places. Yes, I see in #4 you tested, but did you try a different device? Your Android 11 could be the issue and your test device might have an issue with permissions. I don't know.

What the community knows about the Personal Audio Classifier extension . You seem to have found most of the links :slight_smile: One of the links might help with you Q3

It is obviously an imperfect tool that works fairly well for most.

Here is some background for this clever extension https://appinventor.mit.edu/assets/files/Nikhil_Bhatia_MEng_Thesis.pdf

Thank you for the response SteveJG
It is important to me that my app works on a large majority if not all devices so it is better for me to step away from the Personal Classifier and try to achieve desired functionalities some other way.

I am still trying with the Teachable Machine (voice recognition ML model in a form of java script, uploaded as "media" file with html extension) but difficulties with allowing permission for recording and permission to run the java script made me try with "Personal Audio Classifier".

Thank you for your help
Milan