Open GeorgeS2019 opened 5 months ago
As answered in another issue, there are no audio to phonemes for sherpa-onnx. We don't use phonemes at all during model training.
If you train a model by yourself that uses phonemes as the modelling unit inside icefall, then you can use the model in sherpa-onnx.
@csukuangfj https://huggingface.co/bookbot/sherpa-onnx-pruned-transducer-stateless7-streaming-id
Instead of being trained to predict sequences of words, this model was trained to predict sequence of phonemes, e.g. ['p', 'ə', 'r', 'b', 'u', 'a', 't', 'a', 'n', 'ɲ', 'a']. Therefore, the model's vocabulary contains the different IPA phonemes found in g2p ID.
This model was converted from the TorchScript version of Pruned Stateless Zipformer RNN-T Streaming ID to ONNX format.
In that case, you can use it directly with sherpa-onnx. Are there any issues you have when running it with sherpa-onnx?
Is there already an example for this using ONNY