Castilian vs Spanish

[lokesh12345678910]

I've transcribed in Castillian through whisper –model large-v2 –language Castilian. Does this have the same output as --language Spanish?

I asked the owner of a different package to add the Whisper Castilian model to their package. This was their response:

I haven’t found a separate model that supports Castilian Spanish; the appendix of Whisper’s paper (https://cdn.openai.com/papers/whisper.pdf) only lists “Spanish” as a language they had data for.

Perhaps Castilian is an accepted flag that also means Spanish? Have you found that supplying “Castilian” has performance differences on the Whisper implementation you are using?>

Is there a seperate model that supports Castilian?

[glangford]

Is there a seperate model that supports Castilian?

In the current version of whisper, Castilian appears to be aliased to Spanish.

whisper/whisper/tokenizer.py

Line 126 in fcfeaf1

"castilian": "es",

[lokesh12345678910]

Has there been any version of whisper (large-v2 or large-v1) where Castilian isn't aliased to Spanish?

[glangford]

Has there been any version of whisper (large-v2 or large-v1) where Castilian isn't aliased to Spanish?

It looks like the alias to Spanish goes back to the original version of whisper, so I think it has been this way for all models from the beginning.

[sephethus]

Castilian is very distinctly different than American Spanish, more so than British English is from American. I don't understand why this isn't well understood.