fix: initialize encoding variable before conditional in inference

[Mr-Neutr0n]

Summary

• In AcousticModel.inference(), the encoding variable was only assigned inside a conditional block (if encoder_outputs_ex.shape[1] > pos_encoding.shape[1]), meaning it would be undefined when the condition is False.
• This causes a NameError at runtime when self.decoder() is called with encoding=encoding.
• Fixed by defaulting encoding = pos_encoding before the conditional, so the variable is always initialized regardless of whether the expanded encoder output is longer than the original positional encoding.

Test plan

• Verify inference() no longer raises NameError when encoder_outputs_ex.shape[1] <= pos_encoding.shape[1]
• Confirm behavior is unchanged when the condition is True (encoding is still recomputed for the expanded length)

[CLAassistant]

All committers have signed the CLA.

[Mr-Neutr0n]

closing this out — saw the note about the repo being unmaintained. will redirect to the active fork if still relevant