This patch deprecates the language parameter, suggesting users should use a new
model parameter instead.
The idea behind this change is that Vosk provides several different models for
the same language and just using a language code for identification does not
allow us to distinguish between these models. For example, I was surprised to
find out that the packaged English model is the one which is deliberately kept
small for use on mobile devices and does not have a very good error rate.
Loading the language model is now also more flexible sine users can either
specify a path to the model with fallbacks to default locations. That way,
users do not need to write files into system directories for testing or when
they want to use their own model.
This patch deprecates the language parameter, suggesting users should use a new model parameter instead.
The idea behind this change is that Vosk provides several different models for the same language and just using a language code for identification does not allow us to distinguish between these models. For example, I was surprised to find out that the packaged English model is the one which is deliberately kept small for use on mobile devices and does not have a very good error rate.
Loading the language model is now also more flexible sine users can either specify a path to the model with fallbacks to default locations. That way, users do not need to write files into system directories for testing or when they want to use their own model.