Changing hotwords without reloading model

[mxbi]

First of all, thank you for building such a cool library.

I'm using the C++ API to load streaming Zipformer models with SherpaOnnxCreateOnlineRecognizer, using hotwords_file to bias the decoding. In my use case, there are different situations in which we're expecting different utterances, but changing the hotwords appears to require reloading the model (which is obviously)

Is there a way to change the hotwords for an already created recognizer? If not, then I could try to help implement it if you had some pointers on how to do it.

Thanks a lot!

[csukuangfj]

We already support it.

please see

sherpa-onnx/sherpa-onnx/c-api/c-api.h

Line 233 in e424cc9

SherpaOnnxCreateOnlineStreamWithHotwords(

[mxbi]

Thank you @csukuangfj ! Could you clarify how the API should be used?

Specifically, I'm assuming:

1. I shouldn't load the recogniser with hotwords file/score set?
2. const char *hotwords is a path to a hotwords file, rather than a buffer containing the hotwords?