training KeywordRecognizer

[ramkirank]

Can spokestack-python be used independently to train KeywordRecognizer model(or SpokestackKeyword:) . I want to use keras model.fit() or tensorflow training loop.

[will-rice]

The python library does not allow you to directly train models at this time.

If you would like to make a custom model you can do so via Maker with an account.

In addition, you can also check out the docs for more about our wakeword and keyword models.

[ramkirank]

is there any plan to support this feature in future releases? . And can you please suggest where I can find model architecture in detail(like if conv1d or dense layers used in filter/encode/detect.tflite for KeywordRecognizer or others)

[will-rice]

Spokestack is intended as a runtime library so it is unlikely that we would add that feature in the future. Spokestack is intended as a runtime library so it is unlikely that we will support that feature in the future. The keyword model is very similar to the wake word model. The wake word section I linked goes over the basics, but at the bottom of the page, there is a section with references we used in the creation of both models which have a lot more detail.