Currently --similarity_to_quality_model is required for speech mode even though that model file is not used (the exponential mapping is always used for speech).
As a workaround passing in model/libsvm_nu_svr_model.txt as --similarity_to_quality_model will allow running in external directories.
Currently --similarity_to_quality_model is required for speech mode even though that model file is not used (the exponential mapping is always used for speech). As a workaround passing in model/libsvm_nu_svr_model.txt as --similarity_to_quality_model will allow running in external directories.