Is it possible to train a model on phonemes level and use it as streaming on device?

[make1986]

Do you have an example of a pretrained model that I could try before I decide to invest time into it?

[csukuangfj]

You can find a list of pre-trained streaming models at
https://k2-fsa.github.io/sherpa/onnx/pretrained_models/online-transducer/index.html

However, none of them uses phonemes. They use BPE for English.

[ibsusu]

Has this changed since you started converting things with piper-phonemize? Also, thanks for working on this for basically a year and then including espeak-ng -> piper-phonemize into this. It's insane that this is so good and works on so many systems.

The amount of CMake craziness you've gone through is admirable. I have no idea how you got so good at it.