Trouble with Recognition of Words

[hankp46]

I understand that training is a lot of trail & error but am getting random results (recognizes word, recognizes wrong word, doesn't give any result, etc.). My t3 phrases are:

1. Car Forward
2. Car Stop
3. Lights On Here is what my dictionary looks like:

   stop 19f24e40-2605-44fd-864e-42dbf70e5a9e
   lights d96ea764-c979-49d2-8578-40d9a6e501b4
   go 5e208e73-5820-47bb-ad99-15e9e6056505
   go 29da78fa-3281-48ee-af9a-e1f351ea8a32
   go 6e59d3eb-6a35-44ba-a108-09dc6decba78
   stop 28cc4206-402d-4f9d-8c53-4377d404d0dd
   lights 37028bbd-7c53-4368-82cd-a376ae63e83e
   go ff97e064-8198-4f4e-9a6a-7d4c9677985d
   go 352dd769-51ca-4293-9ba4-cb57d936bdf6
   lights 687c92ac-7f0c-4789-99bb-26512ef3c54f
   go b5ec1f63-8694-4645-986e-c1ab0ee72a9c
   go a30545db-c52d-4819-938f-16a078864d99
   lights eb25cf38-533e-4b4f-bcc9-72db1ee10ada
   stop 2a64b369-18a7-4bd3-8ef0-7e9ed0b89e7b
   stop c7d62542-42e1-4013-afbf-131ad944a528
   go 94d8b9a4-a13b-4ed7-a77c-b44fd8ed04f4

   Should the entries look like this or should they be grouped together? I've tried many combination of settings but just can't get one that gives consistent results.

Any help would be appreciated.

[bishoph]

You can can group the output easily with the following command:

./sopare.py -o | tail -n +3 | sort

Make sure that your timings are well defined for your need. Test in a quite room with a relatively low THRESHOLD to catch all input for short words like "go" as it has a smooth first letter. Use the highest sample rate possible and high precision to identify your short words. A good mic is important as well. A potential alternative is to use word combinations. I made a couple of blog post and videos explaining the training process: https://www.bishoph.org