Obtaining Encoder Features

[magacc1]

I'm looking to use some of the pretrained speech recognizers and compute their encoder output features in real-time.

When I'm looking at the API of _Recognizer / _sherpa_onnx.OnlineRecognizer as defined here

I see only the following methods, and cannot find a way to get access to specifically compute the encoder outputs: 'create_stream', 'decode_stream', 'decode_streams', 'get_result', 'is_endpoint', 'is_ready', 'reset'

Can anyone help provide pointers on how to get the encoder features, or find the code underneath decode_stream and break it down to compute the encoder results?

Thanks!

[csukuangfj]

We don't provide such an API as general users don't need it.

Can anyone help provide pointers on how to get the encoder features

For the online transducers, you can take a look at https://github.com/k2-fsa/sherpa-onnx/blob/8af2af84664d3285ba452bf453bb928a3eb6e978/sherpa-onnx/csrc/online-recognizer-transducer-impl.h#L224-L225

[magacc1]

Thanks Fangjun!