Verify alphabet in pb* and tflite models

[JRMeyer]

The alphabet files from Jaco models are inconsistent with the output of the models at runtime. It has been observed that the Jaco Spanish model can produce accented vowels, but the alphabet file does not include them. The alphabet file should be confirmed and uploaded to the zoo for language model generation.

TFModelState::init and TFLiteModelState::init can be modified to print out the loaded alphabet used to train the model here: https://github.com/coqui-ai/STT/blob/653ce25a7ce5bd6cbb564416d847d8afcd5c5e8c/native_client/tfmodelstate.cc#L120

[mariano-balto]

Maybe the above could be the cause of the problem we are seeing on a dockerized ARM environment when using the Jaco models for Spanish with the python (3.9) bindings.

https://github.com/coqui-ai/STT/issues/2284

[zuazo]

The correct alphabet files seem to be the following: https://gitlab.com/Jaco-Assistant/Scribosermo/-/tree/deepspeech/data

As you said, the Spanish alphabet includes accented vowels. Also other language's alphabets like French and Polish.